[refactor](catalog) Catalog spi review 16#65785
Open
morningman wants to merge 62 commits into
Open
Conversation
This multi-month refactor needs persistent state for progress, decisions, risks, and cross-session agent handoff. Establishes a file-based tracking system including dashboard, ADR decision log, deviation log, risk register, per-stage task files, per-connector tracking, and an agent collaboration playbook covering context budget / subagent usage / handoff norms. Closes 18 design decisions (D-001..D-018) and registers 14 risks (R-001..R-014). Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
…T27) (apache#63582) ## Summary Lands the P0 SPI baseline for the catalog-SPI migration (master plan §3.1 / RFC §2.1), with zero impact on the already-migrated JDBC + ES connectors. - **Batch 0** (commits 1-2): SPI types + fe-core bridges — `ConnectorMetaInvalidator`, `ConnectorTransaction`, `ConnectorMvccSnapshot`, `ExternalMetaCacheInvalidator`, `ConnectorMvccSnapshotAdapter`, `PluginDrivenTransactionManager` generalization. - **Batch 1** (commit 3): DDL + Partition SPI — `ConnectorCreateTableRequest` + 4 spec POJOs, 4 new defaults on `ConnectorTableOps`, 3 new fields on `ConnectorPartitionInfo`, fe-core converter, `PluginDrivenExternalCatalog.createTable` routing. - **Batch 2** (commit 4): Import-gate + unit tests — `tools/check-connector-imports.sh` wired through exec-maven-plugin; `FakeConnectorPlugin` covering every default fall-through; routing tests for the invalidator; converter tests for all 4 partition styles + 2 bucket flavors. ## Commits - `[feat](connector) add P0 batch 0 SPI baseline: MetaInvalidator / Transaction / MvccSnapshot` (T03-T08) - `[feat](connector) wire P0 batch 0 SPI into fe-core` (T09-T12) - `[feat](connector) add P0 batch 1 SPI: CreateTableRequest + listPartitions` (T13-T20) - `[feat](connector) add P0 batch 2 gate + unit tests` (T21-T23, T26-T27) ## Test plan - [x] `mvn -pl fe-connector/fe-connector-api,fe-connector/fe-connector-spi -am compile` — SPI modules compile - [x] `mvn -pl fe-core -am compile -Dmaven.build.cache.enabled=false` — fe-core compile - [x] `mvn -pl fe-core checkstyle:check` — 0 violations - [x] `mvn -pl fe-connector validate` — import gate runs and passes (baseline clean) - [x] `mvn -pl fe-core -am test -Dtest='FakeConnectorPluginTest,ExternalMetaCacheInvalidatorTest,CreateTableInfoToConnectorRequestConverterTest,ConnectorPluginManagerTest,ConnectorSessionImplTest'` — 39/39 green - [x] `mvn -pl fe-connector/fe-connector-jdbc,fe-connector/fe-connector-es -am compile` — downstream connectors compile unchanged - [ ] JDBC regression-test suite (T24) — to be exercised by this PR's CI pipeline - [ ] ES regression-test suite (T25) — to be exercised by this PR's CI pipeline ## Tracking Full plan, decisions, and risk log live under `plan-doc/` in the repo (introduced by 6315983, already on the base branch). Per-task status: `plan-doc/tasks/P0-spi-foundation.md`. --------- Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
…pache#63641) ## Summary P1 batch A — close out scan-node SPI consolidation while keeping migration-period fallbacks in place. Three surgical changes route `PluginDrivenExternalTable` first in the nereids translator hot paths so already-migrated SPI connectors (JDBC, ES) take the SPI route, while the existing `instanceof XExternalTable` chains remain as fallbacks for connectors still pending migration (P3–P7). - **T3** — `PhysicalPlanTranslator.visitPhysicalFileScan`: move the existing `PluginDrivenExternalTable` branch from position 8 to position 1; the 7 connector-specific branches (HMS / Iceberg / Paimon / Trino / MaxCompute / LakeSoul / RemoteDoris) stay in place as migration-period fallbacks - **T4** — `PhysicalPlanTranslator.visitPhysicalHudiScan`: add a `PluginDrivenExternalTable` branch routed to `PluginDrivenScanNode.create(...)`, threading `tableSnapshot` + `scanParams` through `FileQueryScanNode` setters; `incrementalRelation` flagged as a P3 Hudi SPI extension TODO. The new branch is unreachable today (`PhysicalHudiScan` is only built for `HMSExternalTable + DLAType.HUDI`), so this is groundwork for P3 with zero current-day runtime impact - **T5** — `LogicalFileScan`: in `computeOutput()`, add a `PluginDrivenExternalTable` branch calling new helper `computePluginDrivenOutput()` — same shape as `computeIcebergOutput`, using `getFullSchema()` + virtualColumns; in `supportPruneNestedColumn()`, add an explicit `PluginDrivenExternalTable → false` branch. Both behaviorally equivalent for JDBC/ES today since they have no hidden cols and no virtualColumns P1 batch B (T1 — delete 13 legacy `Jdbc*Client` + `JdbcFieldSchema`) is deferred to P8 because the 3 fe-core callers — `PostgresResourceValidator`, `StreamingJobUtils`, `CdcStreamTableValuedFunction` — are live CDC streaming code that requires SPI extension for `getPrimaryKeys` / `getColumnsFromJdbc` / `listTables`, which is out of P1 surgical scope. Background and tracking docs live in `plan-doc/` (Master Plan §3.2 P1, tasks/P1-scan-node-cleanup.md, decisions log). ## Test plan - [x] `mvn -pl fe-core -am compile -Dmaven.build.cache.enabled=false` → BUILD SUCCESS - [x] `mvn -pl fe-core checkstyle:check` → 0 violations - [x] JDBC + ES regression-test passing — baseline established in P0 / PR apache#63582 - [ ] PR CI green on this PR - [ ] Manual scan-node smoke for an SPI connector — JDBC `SELECT *` should fall into the new `PluginDrivenExternalTable` branch first 🤖 Generated with [Claude Code](https://claude.com/claude-code) --------- Co-authored-by: Claude Opus 4.7 <noreply@anthropic.com>
…apache#64096) ### What problem does this PR solve? Related PR: apache#63582 (P0 — SPI baseline), apache#63641 (P1 — nereids plugin-driven routing) Problem Summary: This is **P2** of the catalog SPI migration and targets the `branch-catalog-spi` feature branch (continuing P0 apache#63582 and P1 apache#63641). It fully migrates `trino-connector` off the legacy in-tree `fe-core/datasource/trinoconnector/` implementation and onto the connector SPI module `fe-connector-trino`, making `trino-connector` the first connector to complete the SPI consumption playbook that later connectors will reuse as a template. All five batches land together so there is no intermediate state where a newly-created trino catalog cannot be serialized. **Batch A — complete the SPI surface (`fe-connector-trino` only, no fe-core changes)** - `TrinoConnectorProvider.validateProperties`: enforce the required `trino.connector.name` property at `CREATE CATALOG` time (ported from the legacy `checkProperties`). - `TrinoDorisConnector.preCreateValidation`: call `ensureInitialized()` so plugin loading + connector-factory resolution happen at catalog creation instead of being deferred to the first `SELECT`. - `TrinoConnectorDorisMetadata.applyFilter` / `applyProjection`: bridge Trino native filter/projection pushdown, reusing `TrinoPredicateConverter` to translate a Doris `ConnectorExpression` into a Trino `TupleDomain`. `remainingFilter` is conservatively returned as the original expression to match legacy behavior (conjuncts are not stripped; BE re-evaluates them). **Batch B — fe-core bridge for image compatibility** - `GsonUtils`: atomically replace the three legacy `registerSubtype` entries (`TrinoConnectorExternalCatalog` / `Database` / `Table`) with `registerCompatibleSubtype` redirects onto the `PluginDrivenExternal*` hierarchy. This must be atomic — `RuntimeTypeAdapterFactory` rejects duplicate labels, so keeping both bindings would throw at static init. Mirrors what ES/JDBC already did. - `PluginDrivenExternalCatalog.gsonPostProcess`: extract a `legacyLogTypeToCatalogType()` helper that maps `Type.TRINO_CONNECTOR` → `"trino-connector"`; the generic `name().toLowerCase()` would otherwise produce the wrong `"trino_connector"` (underscore) that `CatalogFactory` does not recognize. - `PluginDrivenExternalTable.getEngine()` / `getEngineTableTypeName()`: add `trino-connector` branches that preserve the legacy engine-name / table-type display across `SHOW TABLE STATUS` and `information_schema`. **Batch C — flip the switch** - Add `"trino-connector"` to `CatalogFactory.SPI_READY_TYPES` so catalog creation routes through the SPI path. **Batch D — remove legacy code** - Drop the `instanceof TrinoConnectorExternalTable` scan branch in `PhysicalPlanTranslator` (the `PluginDrivenExternalTable` SPI branch already handles it). - Drop `case "trino-connector"` in `CatalogFactory`. - Delete `fe-core/datasource/trinoconnector/` (10 files) and the now-dead legacy `TrinoConnectorPredicateTest`. - Route the `TRINO_CONNECTOR` db-build case in `ExternalCatalog` to `PluginDrivenExternalDatabase` (mirrors the migrated JDBC case). - **Retained for image compatibility**: the `InitCatalogLog.Type.TRINO_CONNECTOR` and `TableType.TRINO_CONNECTOR_EXTERNAL_TABLE` enums, the GsonUtils redirects, and the `MetastoreProperties` trino-connector entry. **Batch E — tests + tracking docs** - 29 JUnit 5 unit tests over the plugin-free converters: - `TrinoPredicateConverterTest` — `ConnectorExpression` pushdown trees → Trino `TupleDomain` (EQ / range / NE / IN / IS [NOT] NULL / AND / OR, Slice encoding), plus graceful degradation to `TupleDomain.all()` on null/unsupported input. - `TrinoTypeMappingTest` — Trino SPI type → Doris `ConnectorType` (scalars, decimal precision/scale, timestamp precision clamp, array/map/struct, unsupported-type failure). - `TrinoConnectorProviderTest` — `validateProperties` fast-fails when `trino.connector.name` is missing/empty. - No Trino plugin/cluster required; plugin-dependent paths remain covered by the existing `external_table_p0/p2` `trino_connector` regression suites. - Sync the migration tracking docs under `plan-doc/` (already carried on this feature branch since P0). **Net effect**: 28 files, +1025 / −2681 (~1656 LOC net removed). Old FE images holding legacy trino catalogs / databases / tables deserialize onto the `PluginDrivenExternal*` hierarchy through the GsonUtils string-name redirect, with engine-name display preserved. **Deferred (follow-ups, not in this PR)**: - `trino_connector_migration_compat` regression test (old-image deserialization) — requires a running cluster + Trino plugin + docker, unavailable in this dev environment; tracked as a CI/cluster follow-up. - The plugin-install documentation update lives in the `doris-website` repo and is handled separately. ### Release note None ### Check List (For Author) - Test - [x] Unit Test — 29 new tests in `fe-connector-trino` (predicate converter / type mapping / property validation). - [ ] Regression test — existing `trino_connector` suites cover plugin paths; the new old-image compat regression is deferred to a CI/cluster follow-up. - [ ] Manual test (add detailed scripts or steps below) - [ ] No need to test or manual test. Explain why: - [ ] This is a refactor/code format and no logic has been changed. - [ ] Previous test can cover this change. - [ ] No code files have been changed. - [ ] Other reason - Behavior changed: - [x] No. Internal routing moves from the legacy fe-core path to the SPI path; image compatibility, engine-name display, and pushdown semantics all mirror the legacy behavior. All batches land together, so there is no serialization-gap window. - Does this need documentation? - [x] Yes. The trino-connector plugin-install doc update is a follow-up in the `doris-website` repo. ### Check List (For Reviewer who merge this PR) - [ ] Confirm the release note - [ ] Confirm test cases - [ ] Confirm document - [ ] Add branch pick label
…tch design (hybrid, T02-T08) (apache#64143) ## Proposed changes testing with apache#64146 P3 of the catalog-SPI migration (base: `branch-catalog-spi`). Migrates the **hudi** connector following the **hybrid** strategy (D-019): harden the dormant HMS-over-SPI hudi connector to correctness parity, build a test baseline, and write the per-table dispatch design — **all behind the closed gate** (`SPI_READY_TYPES` unchanged). >⚠️ **No user-visible behavior change.** The SPI hudi path stays dormant (gate closed); hudi queries continue to use the legacy `HMSExternalTable.dlaType=HUDI` path. This PR removes correctness blockers ahead of the live cutover (deferred to P7 / batch E). ### What's included **Correctness fixes (hardening dormant code, behind gate):** - **T02** — fix hudi JNI `column_types` double bug: emit full Hive type strings (was Doris bare type names, losing precision/scale/subtypes) and send `column_names`/`column_types`/`delta_logs` as typed lists end-to-end (was comma join/split, which shattered `decimal(10,2)` / `struct<...>`). Matches the BE `hudi_jni_reader.cpp` contract (names `,` / types `#` / delta `,`). - **T04** — fail loud on time-travel / incremental read in the SPI `visitPhysicalHudiScan` branch (was silently returning the latest snapshot / silently full-scanning). - **T05** — real EQ/IN partition pruning in `HudiConnectorMetadata.applyFilter` (was a placeholder that ignored predicates and unconditionally switched the partition source from Hudi-metadata to HMS); faithfully mirrors `HiveConnectorMetadata.applyFilter`. - **T07** — column-name casing fix in `avroSchemaToColumns` (top-level lowercase, mirroring legacy `HMSExternalTable`). **Test baseline (all three connector modules started P3 with 0 tests):** - `fe-connector-hudi` (33): type-mapping / schema-parity (COW/MOR golden) / table-type / partition-pruning / scan-range. - `fe-connector-hms` (12): shared Hive-type-string parser tests. - `fe-connector-hive` (14): file-format / partition-pruning (mirrors T05). - COW/MOR schema is **type-agnostic** (golden parity vs legacy `initHudiSchema`); table type only affects scan planning. **Decisions / design (code-grounded, design-only):** - **T03** — defer `schema_id`/`history_schema_info` field-id evolution to batch E (DV-006; not a model-agnostic SPI fix). - **T06** — keep MVCC/snapshot SPI defaults (opt-out) + document (DV-007). - **T08** — `tableFormatType` dispatch design memo + **D-020**: single `hms` catalog per-table routing via a new backward-compatible `ConnectorMetadata.getScanPlanProvider(handle)` (per-table provider seam); refines D-005. The keystone gap is split into M1 (identity consumption, fe-core reads `tableFormatType` as an opaque string) and M2 (scan routing). ### Deferred to batch E / P7 (not in this PR) Gate flip (`SPI_READY_TYPES += hms/hudi`), fe-core `tableFormatType` consumption (M1+M2 implementation), live cutover, delete legacy `datasource/hudi/`, full incremental/time-travel/MVCC, Iceberg-on-hms via SPI (needs P6 `IcebergScanPlanProvider`), cluster/runtime validation. ### Verification Per task tracking, each code batch landed with: per-module compile + checkstyle 0 (incl. test sources) + connector import-gate pass + new unit tests green. The two most recent commits are docs-only (`plan-doc/`); the code is unchanged since the last green batch. Gate stays closed → the dormant SPI path is unreachable at runtime → zero live-path risk. CI re-verifies. 🤖 Generated with [Claude Code](https://claude.com/claude-code)
…core + make fe-core odps-free (T07-T09) (apache#64300) Follow-up to apache#64253 (the MaxCompute catalog-SPI cutover). After the cutover a `max_compute` catalog deserializes to `PluginDrivenExternalCatalog` and no legacy `MaxComputeExternal*` object is ever instantiated, so the legacy MaxCompute subsystem in fe-core is dead code. This removes it and makes fe-core's dependency tree fully odps-free. **1. Remove legacy subsystem** (`7a4db351100`) - Delete 20 fe-core files: `datasource/maxcompute/*` (incl. `MCTransaction`, `MaxComputeScanNode`/`Split`), the MaxCompute sink/insert/txn plumbing, and 2 legacy-only tests. - Clean ~21 reverse-reference sites (imports + dead `instanceof`/visitor/rule branches), keeping every `PluginDriven`/connector sibling branch and the image/replay keep-set (GsonUtils compat strings; `TableType`/`TransactionType`/`TableFormatType`/`InitCatalogLog.Type` `MAX_COMPUTE` enums; block-id thrift). - Rewire 3 tests; e.g. `FrontendServiceImplTest`'s block-id RPC test now mocks the generic `Transaction` SPI, since `getMaxComputeBlockIdRange` reads the PluginDriven connector transaction. **2. Make fe-core odps-free** (`409300a75b8`) - Drop the two odps deps from `fe-core/pom.xml`. - Move `MCUtils` from fe-common into `be-java-extensions/max-compute-connector` (its only consumer after the removal); keep `MCProperties` (odps-free constants) in fe-common. - Drop `odps-sdk-core` from fe-common — it was also leaking netty/protobuf transitively to fe-common's own `DorisHttpException`/`GsonUtilsBase`, so declare `netty-all` + `protobuf-java` directly (proper dependency hygiene). **3. Doc-sync** (`f8c305765e8`) — plan-doc PROGRESS/HANDOFF/deviations/design tracking notes. - `mvn -pl :fe-core -am test-compile` (main+test) passes; checkstyle 0 violations; connector import-gate passes. - `grep -rn com.aliyun.odps fe/fe-core/src` → empty. - `mvn -pl :fe-core dependency:tree | grep odps` → empty (no odps, direct or transitive). 🤖 Generated with [Claude Code](https://claude.com/claude-code) --------- Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
…he#64446) (apache#64446) Migrates the legacy in-tree Paimon catalog (fe-core `datasource/paimon`, `metacache/paimon`, `property/metastore/*Paimon*` and their reverse references) onto the catalog SPI as a self-contained `fe-connector-paimon` plugin, following the maxcompute full-adopter + cutover template. `paimon` is added to `SPI_READY_TYPES`, so a Paimon catalog now deserializes to `PluginDrivenExternalCatalog` and all five functional areas — normal-table read, system-table read, DDL, MTMV/MVCC/time-travel (procedures are a doc-only no-op: fe-core has none) — flow through the generic `PluginDriven*` bridge with no Paimon-specific branch in fe-core. Paimon is the first lakehouse full-adopter to exercise the MVCC / sys-table / MTMV SPI surfaces, which become the reusable template for the upcoming iceberg/hudi migrations, so the cutover is held to byte-parity with the legacy path. The legacy `datasource/paimon/*` classes are left in place (now dead on the cutover path) and removed in a follow-up, mirroring the maxcompute cutover -> P4 removal. DefaultConnectorContext) Source-agnostic seams so any connector can express these capabilities; every new method is a default no-op/identity, so hive/iceberg/jdbc/maxcompute/ trino/es are byte-for-byte unaffected. - MVCC/time-travel: `resolveTimeTravel(session, handle, ConnectorTimeTravelSpec)` + `applySnapshot(...)` (threads a pinned snapshot into the handle before planScan); new immutable `ConnectorTimeTravelSpec` (SNAPSHOT_ID / TIMESTAMP / TAG / BRANCH / INCREMENTAL) that fe-core extracts from SQL; `ConnectorMvccSnapshot` carries `schemaId` for schema-evolution-aware time travel. - System tables: `listSupportedSysTables` / `getSysTableHandle`; fe-core composes the `{base}${sys}` reference name. - Scan: proportional split-weight (`getSelfSplitWeight`/`getTargetSplitSize`), COUNT(*) pushdown (`getPushDownRowCount`, 7-arg `planScan`), native-read accounting (`isNativeReadRange`), `getDeleteFiles` (VERBOSE merge-on-read), `ignorePartitionPruneShortCircuit` (predicate-driven connectors keep genuine-null partitions). - Metadata: `ConnectorPartitionInfo.fileCount` + `SUPPORTS_PARTITION_STATS` (rich SHOW PARTITIONS); `ConnectorColumn.withTimeZone` (WITH_TIMEZONE DESC marker, independent of the timestamp_tz mapping flag); connector cache control (`invalidateTable`/`invalidateAll`, `schemaCacheTtlSecondOverride`). - `ConnectorContext`: six engine-side hooks the connector must not implement itself (`loadHiveConfResources`, `vendStorageCredentials`, `getBackendStorageProperties`, `normalizeStorageUri` x2, `getStorageProperties`), bound in `DefaultConnectorContext` over fe-core's single-source-of-truth credential/URI/HiveConf utilities (no re-ported logic that could drift). The metastore-side counterpart of the fe-filesystem StorageProperties SPI: a Paimon catalog resolves its backend (hive/dlf/rest/jdbc/filesystem) by ServiceLoader dispatch on `paimon.catalog.type` instead of a fe-core switch. - `fe-connector-metastore-api`: hadoop/SDK-free `MetaStoreProperties` contracts (HMS/DLF/REST/JDBC/FileSystem) carrying only Map/scalar facts. - `fe-connector-metastore-spi`: `MetaStoreProvider<P>` + first-hit `MetaStoreProviders` dispatcher + five impls, registered via META-INF/services. `HmsMetaStorePropertiesImpl` reproduces the legacy `buildHmsHiveConf` key set and parity-critical ordering (storage overlay before the kerberos block; username/sasl last). Adding a backend = one provider class + one services line (no central enum/switch). - ServiceLoader is loaded with the SPI interface's own (plugin) classloader, NOT the thread-context CL: at CREATE CATALOG the static init first fires on an FE worker thread whose TCCL is the FE app loader, so a 1-arg load would cache an empty provider list process-wide. - fe-core bridges add `initExecutionAuthenticator(List<StorageProperties>)` so HDFS-Kerberos doAs is wired on the plugin path (legacy `initializeCatalog` is dead). Full read + DDL + partitions + statistics + system tables + MVCC/time-travel across all five metastore flavors, planning native (ORC/Parquet) vs JNI reads. - `PaimonCatalogFactory` (pure flavor switch: Options + Hadoop Configuration + HiveConf, classloader-pinned for child-first loading), `PaimonCatalogOps` (injection seam over the remote Catalog for offline-testable metadata), `PaimonConnectorMetadata` (~9 -> 28 methods: schema latest/at-snapshot, sys tables, MVCC/time-travel, DDL, partitions, statistics, table descriptor). - `PaimonScanPlanProvider` (~6 -> 30 methods): predicate/projection pushdown, native-vs-JNI routing, large-file sub-splitting with deletion vectors, COUNT(*) collapse, schema-evolution field-id dictionary, `$ro` unwrap to base FileStoreTable. - New: `PaimonSchemaBuilder` (CREATE TABLE), `PaimonIncrementalScanParams` (@incr), `PaimonTableResolver` (sys/branch-aware handle -> Table), `PaimonLatestSnapshotCache` + `PaimonSchemaAtMemo` (per-catalog caches on the long-lived connector). - `PaimonTypeMapping` gains the write direction (Doris -> Paimon) for CREATE TABLE. - NEW `PluginDrivenMvccExternalTable` (generic MvccTable + MTMV host: the runtime class for paimon tables), `PluginDrivenMvccSnapshot`, `PluginDrivenSysExternalTable` + `systable/PluginDrivenSysTable`. Behavior is selected by `ConnectorCapability` (SUPPORTS_MVCC_SNAPSHOT, SUPPORTS_PARTITION_STATS) — no paimon branch in fe-core. - `PluginDrivenScanNode`: MVCC pin threaded at every handle-consumption site, sys-table time-travel guard, COUNT pushdown, connector-agnostic EXPLAIN re-emission. - Cutover: `CatalogFactory` adds `paimon` to `SPI_READY_TYPES` and drops the built-in case; `GsonUtils` removes the 7 built-in `Paimon*` subtype registrations and adds replay compat so persisted `PaimonExternalCatalog` (all 5 flavors) / `...Database` -> `PluginDriven*` and `PaimonExternalTable` -> `PluginDrivenMvccExternalTable` on deserialization (existing clusters upgrade without losing catalogs). - Nereids/DDL: `PhysicalPlanTranslator` drops the `PaimonScanNode` branch; `UserAuthentication`, `CreateTableInfo`, `ShowPartitionsCommand` generalized; `Env.getDdlStmt` renders LOCATION/PROPERTIES for SHOW CREATE (paimon engine, no credential leak). - fe-filesystem: NEW `HdfsFileSystemProperties` / `HdfsConfigFileLoader` (typed HDFS; defaults-laden BE map vs defaults-free FE Hadoop-config map so it cannot clobber a co-bound object store's fs.s3a.* during a multi-backend merge); S3 MinIO support + legacy tuning defaults; OSS/OBS/COS ANONYMOUS provider-type; new `FileSystemFactory.bindAllStorageProperties` / `FileSystemPluginManager.bindAll`. - NEW `fe-kerberos` leaf module: dependency-free `AuthType` / `KerberosAuthSpec` facts shared across modules without pulling in Hadoop. - Packaging: NEW `fe-connector-paimon-hive-shade` relocates `org.apache.thrift` for HMS 2.3.7; the paimon plugin-zip excludes fe-thrift/libthrift (provided parent-first) to avoid a split `TBase` across loaders; self-contained bundle (hadoop-aws, AWS-SDK, hadoop-hdfs-client). `paimon.version` (1.3.1) pinned as the single property across FE connector / BE paimon-scanner / preload-extensions for FE<->BE Table/Split serde. - BE `PaimonJniScanner.getPredicates()` null-predicate backstop (no-filter scan; pairs with the FE producer now always emitting an empty predicate list). Parity fixes surfaced by the SPI cutover (each restores legacy semantics): - Storage/credentials: canonical `s3.*`/`oss.*` keys -> `fs.s3a.*`/`fs.oss.*` (STORAGE-CREDS); canonical `AWS_*` BE creds (STATIC-CREDS-BE); per-table vended DLF/STS token (REST-VENDED); `oss/cos/obs/s3a` -> `s3://` for data + deletion-vector paths (URI-NORMALIZE); external `hive.conf.resources` hive-site.xml (HMS-CONFRES); real doAs on filesystem/jdbc + wrapped HMS read RPCs (KERBEROS-DOAS). - Reads/types: paimon columns forced nullable (READ-NOTNULL); `VARCHAR(65533)` not widened to STRING (VARCHAR-BOUNDARY); dotted `enable.mapping.varbinary` / `.timestamp_tz` keys (MAPPING-FLAG-KEYS); per-type partition-value rendering (NATIVE-PARTVAL); genuine `\N` not coerced to NULL (PARTITION-NULL-SENTINEL); field-id history dict for rename-safe native reads (SCHEMA-EVOLUTION); paimon-native split encode under `enable_paimon_cpp_reader` (CPP-READER). - Planning/perf: restore the `force_jni_scanner` escape hatch (FORCE-JNI-SCANNER); intra-file native sub-splitting (NATIVE-SUBSPLIT); precomputed COUNT(*) row count (COUNT-PUSHDOWN); table row count for CBO / SHOW TABLE STATUS (TABLE-STATS). - Time travel / DDL: `FOR TIME AS OF` under CST/PST session zones (TZ-ALIAS); resolved + validated `driver_url` for jdbc catalogs (JDBC-DRIVER-URL); reject a local-only name conflict on CREATE TABLE (CREATE-TABLE-LOCAL-CONFLICT). plan-doc P5 design / task-list / HANDOFF / deviation tracking notes. - ~70 new FE unit suites: paimon connector (offline recording fakes: `RecordingPaimonCatalogOps` / `FakePaimonTable`), fe-core PluginDriven* MVCC / sys-table / scan-node / split-weight, metastore-spi dispatch + per-backend properties, fe-filesystem HDFS/MinIO/anonymous, fe-kerberos. - checkstyle 0 violations; connector import-gate passes; FE main+test compile clean. - End-to-end paimon behavior is docker-gated (`enablePaimonTest=true`); the five functional areas are covered by the `external_table_p0/paimon` regression suites, green on the latest CI run for this branch.
… make fe-core paimon-SDK-free (T29) (apache#64653) Follow-up to apache#64446 (the Paimon catalog-SPI cutover). After the cutover a `paimon` catalog deserializes to `PluginDrivenExternalCatalog` and its tables to `PluginDrivenMvccExternalTable`, so no legacy `Paimon*` catalog / table / scan-node object is ever instantiated — the entire legacy Paimon subsystem in fe-core is dead code. This PR removes it and makes fe-core's dependency tree fully paimon-SDK-free (zero `org.apache.paimon` imports, the 5 paimon maven deps dropped). Mirrors the P4 maxcompute cutover → legacy-removal (apache#64256). **1. Remove legacy subsystem + sever reverse-refs** (`7632a074e4b`) - Delete the dead fe-core paimon files: `datasource/paimon/*` (all 5 flavor catalogs, `PaimonExternalTable`, `PaimonMetadataOps`, `PaimonUtil`, `source/PaimonScanNode` / `PaimonSplit` / `PaimonPredicateConverter` / `PaimonSource` / `PaimonValueConverter`, `profile/*`), `datasource/metacache/paimon/*`, `systable/PaimonSysTable`, and the legacy-only tests. - Sever the live reverse-references to the deleted classes, keeping every `PluginDriven` / connector sibling branch and the image/replay keep-set: drop the `PAIMON` db-creation switch arm in `ExternalCatalog`; the paimon engine routing in `ExternalMetaCacheMgr` / `ExternalMetaCacheRouteResolver`; the dead `PAIMON_EXTERNAL_TABLE` branch in `Env.getDdlStmt` (keep the LIVE `PLUGIN_EXTERNAL_TABLE` SHOW CREATE LOCATION/PROPERTIES rendering); the `PaimonSysExternalTable` branch in `UserAuthentication`; the legacy clauses + `handleShowPaimonTablePartitions()` in `ShowPartitionsCommand` (keep `hasPartitionStatsCapability()` and the `TableType.PAIMON_EXTERNAL_TABLE` enum). Landed as one compiling unit (mirrors the P4 apache#64300 precedent: `PaimonUtils` still called the removed `ExternalMetaCacheMgr.paimon()`). - Inline the `getPaimonCatalogType()` string literals (`hms`/`filesystem`/`dlf`/`rest`/ `jdbc`) to decouple the still-consumed thin metastore-property classes from the deleted `PaimonExternalCatalog`. **2. Make fe-core paimon-SDK-free** (`32de7d16702`) - Strip the dead paimon-SDK catalog-building cluster from `AbstractPaimonProperties` + the 5 flavors (`initializeCatalog` / `buildCatalogOptions` / `appendCatalogOptions` / `getMetastoreType` / `normalizeS3Config` / … — 0 live callers; the plugin path builds catalogs connector-side). Keep every LIVE SDK-free duty: `warehouse` `@ConnectorProperty`, the Kerberos `executionAuthenticator` doAs wiring (read by `PluginDrivenExternalCatalog`), property normalization/validation, `getPaimonCatalogType`, `Type.PAIMON`. - Vended credentials: delete `PaimonVendedCredentialsProvider` + its test + the `VendedCredentialsFactory` `case PAIMON`. Its SDK methods were dead — reachable only via the iceberg-only `getStoragePropertiesMapWithVendedCredentials`; the real paimon vended path is the connector's `PaimonScanPlanProvider.extractVendedToken` (moved off fe-core at cutover). Relocate its one LIVE duty (the REST "skip static storage map" gate) to a new SDK-free `MetastoreProperties.isVendedCredentialsEnabled()` (base=false, `PaimonRestMetaStoreProperties`=true); `CatalogProperty` routes iceberg through its provider (byte-identical) and everything else through the metastore-props method. - pom: drop `paimon-core` / `-common` / `-format` / `-s3` / `-jindo` from `fe-core/pom.xml`. `fe/pom.xml` `paimon.version` is kept (fe-connector-paimon + BE paimon-scanner still consume it). **3. Doc-sync** (`58160cbcb56`, `895e214b2bd`) — plan-doc HANDOFF / PROGRESS / connectors / P5-T29 legacy-removal design tracking notes. - `mvn -pl :fe-core -am test-compile` (main + test) passes; checkstyle 0 violations; connector import-gate (`tools/check-connector-imports.sh`) passes. - `grep -rn org.apache.paimon fe/fe-core/src` → empty (main and test). - `mvn -pl :fe-core dependency:tree -Dincludes=org.apache.paimon` → empty (no paimon dep, direct or transitive). - End-to-end paimon regression (`external_table_p0/paimon`, docker `enablePaimonTest=true`) passed on this branch.
…kerberos (apache#64655) **P3b: consolidate the drifted Kerberos/Hadoop authentication implementations into the new top-level neutral leaf module `fe-kerberos`** as the single source of truth. Done as 3 commits: 1. **trino → JDK** (`4a740e1`) — replace the only external dependency in the auth path, trino's `KerberosTicketUtils`, with a JDK-only (`javax.security.auth.kerberos`) byte-for-byte equivalent, so the kerberos path is trino-free. 2. **relocate** (`8898e15`) — move the 13 `fe-common` `security.authentication.*` classes to `org.apache.doris.kerberos.*` in `fe-kerberos`; retarget all consumer imports (fe-core + 3 be-java-extensions scanners); merge the duplicate `AuthType`. 3. **unify interface** (`5e3e896`) — merge the two competing `HadoopAuthenticator` interfaces (fe-common's `PrivilegedExceptionAction` variant vs fe-filesystem-spi's `IOCallable` variant) into the single fe-kerberos one, and delete fe-filesystem-hdfs's own `KerberosHadoopAuthenticator`/`SimpleHadoopAuthenticator` copies (which had drifted from the canonical impls). `DFSFileSystem` now routes through the shared authenticators. `fe-kerberos` remains a top-level neutral leaf (no dependency cycle). HDFS filesystem access now uses the same authenticators as the HMS path (restoring parity). Two intentional behavior changes in fe-filesystem-hdfs: simple / no-`hadoop.username` now runs as remote user `hadoop` (was: FE process user, direct); kerberos uses the shared `LoginContext` + 80%-lifetime refresh. fe-filesystem-hdfs 79/0/0 (+fe-kerberos/spi), checkstyle 0, connector import-gate clean, whole-repo grep for the removed symbols = 0. >⚠️ docker kerberos e2e (HDFS kerberized + HMS) NOT yet run — the real gate; UGI login can't be exercised in unit tests.
…move legacy fe-core subsystem (apache#64688) Migrates the legacy in-tree Iceberg catalog (fe-core `datasource/iceberg`, its `property/metastore/*Iceberg*` clusters, and every reverse `instanceof IcebergExternal*` / `case ICEBERG` coupling scattered across nereids / planner / alter) onto the catalog SPI as a self-contained `fe-connector-iceberg` plugin, following the paimon full-adopter + cutover template. `iceberg` is added to `SPI_READY_TYPES`, so an Iceberg catalog now deserializes to `PluginDrivenExternalCatalog` and every functional area — read, write, row-level DML, `ALTER TABLE EXECUTE` procedures, system tables, DDL and MVCC/time-travel — flows through the generic `PluginDriven*` bridge with no Iceberg-specific branch in fe-core. (Iceberg tables surfaced *through* a `hive`/HMS catalog stay on the hive path until P7; this PR covers the dedicated `type=iceberg` catalog.) Iceberg is the widest lakehouse adopter migrated so far — 7 catalog flavors, merge-on-read row-level DML, 9 stored procedures, v3 row-lineage / deletion vectors — so the cutover is held to behavior-parity with the legacy path. Unlike the paimon migration (a cutover PR + a separate removal PR), this PR does the build-out, the live cutover **and** the full legacy-subsystem removal on one branch: fe-core `datasource/iceberg` and its property clusters are deleted, not just left dead. Part of the catalog-SPI migration tracked in apache#65185 (phase **P6**). Full read + write + row-level DML + procedures + system tables + DDL + MVCC/time-travel, planning native (Parquet/ORC) vs JNI reads. Key seams: - `IcebergCatalogFactory` (pure flavor switch, classloader-pinned for child-first loading), `IcebergCatalogOps` (metadata / DDL ops), `IcebergConnectorMetadata` (schema latest / at-snapshot, sys tables, MVCC/time-travel, DDL, partitions, statistics, table descriptor), `IcebergConnectorProvider` (ServiceLoader entry). - `IcebergScanPlanProvider` + `IcebergScanRange`: predicate/projection pushdown (`IcebergPredicateConverter` → Iceberg `Expression`), FileScanTask byte-offset sub-splitting, COUNT(\*) collapse, merge-on-read delete attachment (position delete / Puffin DV / equality delete), `IcebergColumnHandle` field-id column pruning, schema-evolution-safe reads (`IcebergSchemaUtils` field-id dictionary), synthetic rowid + v3 row-lineage metadata columns. - Write path: `IcebergWritePlanProvider` / `IcebergWriteContext` / `IcebergWriterHelper` / `IcebergConnectorTransaction` — INSERT / INSERT OVERWRITE (dynamic + static partition), branch-targeted writes, WRITE ORDERED BY, per-spec distribution, optimistic-concurrency conflict detection, one SDK transaction per statement, BE commit-fragment → DataFile/DeleteFile conversion. - Row-level DML: DELETE / UPDATE / MERGE INTO over position-delete (v2) and deletion-vector (v3) merge-on-read, baseSnapshot-anchored read. - `action/`: the 9 `ALTER TABLE EXECUTE` procedures (rollback_to_snapshot, rollback_to_timestamp, set_current_snapshot, cherrypick_snapshot, fast_forward, expire_snapshots, publish_changes, rewrite_manifests, rewrite_data_files) dispatched by `IcebergExecuteActionFactory`; `rewrite/` hosts the distributed bin-pack `rewrite_data_files` engine. - `IcebergSchemaBuilder` (CREATE TABLE), `IcebergTypeMapping` (both directions), `IcebergColumnChange` / `IcebergComplexTypeDiff` (schema + partition evolution), `IcebergLatestSnapshotCache` / `IcebergManifestCache` (per-catalog caches on the long-lived connector), `dlf/` (Aliyun DLF client). (`fe-connector-metastore-iceberg`) An Iceberg catalog resolves its backend by `MetaStoreProvider` ServiceLoader dispatch on `iceberg.catalog.type` (registered via `META-INF/services`), instead of a fe-core switch — **7 flavors**: REST, Hive Metastore, AWS Glue, Hadoop / filesystem, JDBC, Aliyun DLF, AWS S3Tables. Adding a backend = one provider class + one services line. The connector self-builds the HMS-metastore Kerberos authenticator, so metastore auth is owned entirely by the plugin. - `CatalogFactory` adds `iceberg` to `SPI_READY_TYPES` and drops the built-in case. - `GsonUtils` replay-compat: persisted `IcebergExternalCatalog` → `PluginDrivenExternalCatalog`, `IcebergExternalDatabase` → `PluginDrivenExternalDatabase`, `IcebergExternalTable` → `PluginDrivenMvccExternalTable` on deserialization, so existing clusters upgrade without losing catalogs (metadata backward-compat invariant). - Nereids/DDL/alter generalized off the concrete Iceberg types; SHOW CREATE renders LOCATION / PARTITION BY / ORDER BY / PROPERTIES via the generic path (iceberg engine name preserved, no credential leak). Per the 2026-07 architecture decision (fe-core holds **no** property parsing for migrated connectors; storage parsing → `fe-filesystem`, metastore parsing → `fe-connector`, both plugin-side; the plugin assembles → BE thrift → hands back to fe-core), iceberg's property/credential/auth resolution moves entirely into the connector: - Storage credentials bound to `fe-filesystem` on both scan and write paths (one source), retiring the second fe-core `StorageProperties.createAll` parse. - Vended temporary credentials, `warehouse` → `fs.defaultFS` derivation (hadoop flavor only, via the neutral `Connector.deriveStorageProperties` SPI), and Kerberos doAs are all connector-owned; the fe-core pre-execution authenticator is a no-op for plugin catalogs. - fe-core's shared static property/metastore helpers are kept alive only for the connectors still on the legacy path (Hive / Hudi / LakeSoul / HMS-iceberg). Net **−21.7k** lines from fe-core. Removes the 4 dead entity classes (`IcebergExternalTable` / `IcebergExternalDatabase` / `IcebergSysExternalTable` / `IcebergExternalCatalog`), the fe-core Iceberg `MetastoreProperties` cluster + its credential helpers, the 5 fe-core Iceberg connectivity probes and the coordinator branches, and un-registers the Iceberg property factories (fail-loud on any stray call). The now-dead paimon `MetastoreProperties` cluster left behind by P5 is swept in the same pass. `GsonUtils` retains only the three string compat labels for old-image upgrade. Parity fixes surfaced by the SPI cutover (each restores legacy semantics): - Classloader: TCCL pinned to the plugin classloader at all three loci where fe-core crosses into bundled Iceberg (scan call thread, write/DDL/commit engine thread via `TcclPinningConnectorContext`, and Iceberg's own manifest-writer worker pool), preventing a duplicate-copy `ClassCastException`. - Nested schema dictionary field names lowercased so a mixed-case nested struct (e.g. Iceberg `DROP_AND_ADD`) can't crash BE with `std::out_of_range`. - Session time zone: lazy `ZoneId` resolution tolerant of Doris CST/PST-style aliases in `FOR TIME AS OF` and datetime partition predicates. - Storage / URI normalization (`oss/cos/obs/s3a` → `s3`), SHOW CREATE secret redaction, and `ALTER TABLE EXECUTE ... WHERE` rejection kept at legacy wording. - ~68 FE unit test classes: iceberg connector (offline recording fakes for the remote Catalog / Table), metastore-iceberg per-flavor dispatch + properties, and the fe-core `PluginDriven*` MVCC / sys-table / scan-node / GSON-replay surfaces; many carry mutation checks. - checkstyle 0 violations; connector import-gate passes (no `fe-connector-iceberg → fe-core` import); FE main + test compile clean. - End-to-end iceberg behavior is docker-gated (`enableIcebergTest=true`); read, write, row-level DML, procedures, system tables, DDL and time-travel are covered by the `external_table_p0/iceberg` (+ `external_table_p2/iceberg`) regression suites. --------- Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
…etaspace leak JdbcConnectorClient cached JDBC-driver classloaders in a ref-counted map that evicted the entry when refCount reached 0. A JDBC driver self-registers into the static java.sql.DriverManager on class load, and close() never deregisters it, so an evicted classloader stays pinned by DriverManager for the life of the process. Serial CREATE -> DROP -> CREATE churn of the same driver URL therefore rebuilt a fresh, separately-pinned classloader every cycle, leaking one driver's worth of Metaspace per cycle. Live FE evidence (external regression build 986696 OOM): only 4 jdbc catalogs yet 38 FactoryURLClassLoader instances survived a full GC (jmap -histo:live), tracking the DriverInfo count -> ~34 leaked, DriverManager-pinned driver classloaders. FE Metaspace grew 165MB -> 1565MB over the single-concurrency 5.5h run (multi-concurrency did not OOM: overlapping same-URL catalogs keep refCount > 0 so the loader was reused instead of evicted-and-rebuilt). Fix: revert to a keep-alive Map<URL, ClassLoader> (one loader per distinct driver URL, shared across catalogs, never evicted), mirroring the pre-SPI fe-core JdbcClient and the pattern IcebergConnector / PaimonConnector already use. Bounded by the number of distinct driver jars. Adds a regression test asserting repeated create/recreate cycles for one URL add exactly one cached classloader. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01XKf5EzHW75qzgrXnkoBZcF
…TableTest after rebase Rebase-integration fix, belongs with P4 (apache#64300 "remove legacy maxcompute subsystem"); squash into it when finalizing that PR. Upstream apache#64937 added MaxComputeExternalTableTest as a unit test for the legacy MaxComputeExternalTable.parsePartitionValues. P4 deletes that class, but the new test file rode into the rebase base with no modify/delete conflict and then dangled: fe-core test-compile failed with "cannot find symbol: MaxComputeExternalTable". No coverage lost: the new connector's MaxComputeConnectorMetadata.listPartitionValues already resolves partition values by name via ODPS PartitionSpec.get(column) (structurally immune to the out-of-order/invalid-spec bug apache#64937 fixed in the legacy path). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
…es into SPI connectors After rebasing branch-catalog-spi onto upstream master (4dfcb4f), an audit of the 3 upstream commits touching fe-core datasource/ found behaviour that was dropped when P5/P6 removed the legacy fe-core paimon/iceberg subsystems. Port the bounded correctness gaps into the new fe-connector SPI code (main + tests): - apache#65332 (paimon JNI IOManager): PaimonScanPlanProvider.getBackendPaimonOptions returned emptyMap() for non-jdbc catalogs, so the three JNI IOManager options (paimon.doris.enable_jni_io_manager / .tmp_dir / .impl_class) never reached BE and Paimon primary-key merge reads could not spill (OOM risk). Collect the three options (strip the "paimon." prefix) before the metastore-type check and forward them for all catalog flavours; jdbc-only driver logic stays in the jdbc branch. - apache#65094 tier1 (build-time correctness): IcebergSchemaBuilder buildPartitionSpec / buildSortOrder and PaimonSchemaBuilder primary/partition keys used bare column names, so a partition/sort/key referenced with a different case than the (now case-preserving) schema column failed the engine's case-sensitive lookup. Add case-insensitive resolution back to the canonical column name (iceberg via Schema.caseInsensitiveFindField; paimon via a lower->canonical map), mirroring upstream getIcebergColumnName / getPaimonColumnNames. - apache#65094 tier2 (consistency, worse than upstream): PaimonConnectorMetadata emitted "partition_columns" with case-preserving keys while column names are lowercased, so PluginDrivenExternalTable's case-sensitive byName lookup silently dropped mixed-case paimon partition columns (table treated as non-partitioned). Lower-case partition_columns with the same bare toLowerCase() the columns use. Verified: mvn package -pl :fe-connector-paimon -am (95 paimon tests) + iceberg 23 tests, 0 failures. Read-path column-name case preservation (tier3) and the apache#63068 OIDC session-credential port are tracked separately. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_013KKXx3btbQc6w3MbZbEoRf
…column-name case on the read path Align the iceberg + paimon catalog-SPI connectors with upstream apache#65094: external table top-level column names now keep their remote case on the read path instead of being lowercased. FE-only (BE untouched): the byte-match invariant "Doris scan slot name == TSchema top-level TField name" is preserved by flipping BOTH the scan slot names AND the field-id-dict / current-TSchema top-level names to case-preserving together, so BE (which matches top-level by field-id) needs no change. Nested struct children stay as-is (iceberg lowercase for BE StructNode.children.at; paimon paimon-cased). Reverts tier2's lowercasing of paimon partition_columns (bce28b2) now that the paimon columns are themselves case-preserving. iceberg (IcebergConnectorMetadata): partition_columns / getColumnHandles / parseSchema. iceberg (IcebergPartitionUtils): getIdentityPartitionColumns (path_partition_keys), getIdentityPartitionInfoMap (scan partition-value map) and generateRawPartition -- the listPartitions value-map key that PluginDrivenExternalTable.getNameToPartitionItems looks up by the case-preserved partition_columns remote name; omitting it would silently drop a mixed-case partition column's value in MTMV / partition pruning. paimon (PaimonConnectorMetadata): partition_columns / getColumnHandles / mapFields. paimon (PaimonScanPlanProvider): path_partition_keys / getPartitionInfoMap and the current (-1) TSchema entry (buildSchemaInfo lowercaseTopLevelNames -> false at the sole current-schema call site; the historical entries and nested names are unchanged). Tests: flip the connector tests that pinned mixed-case -> lowercase output to assert case preservation; add an IcebergSchemaUtils mixed-case byte-match case and an IcebergPartitionUtils.listPartitions regression pinning the case-preserved partition-value-map key. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_013KKXx3btbQc6w3MbZbEoRf
…l DTO + user-session capability Re-migrate the Iceberg REST OIDC per-user session-credential feature (upstream apache#63068 / e545f1a, dropped by the 2026-07-09 P6 rebase) onto the catalog SPI. This is the SPI layer (fe-connector-api): - ConnectorDelegatedCredential: neutral, immutable DTO carrying a user's per-connection OIDC/JWT/SAML token across the fe-core -> connector boundary so the connector never imports a fe-core type. The token is connection-scoped and in-memory only; toString() redacts it (never edit-logged / persisted / SHOWn). - ConnectorSession.getDelegatedCredential() (default empty) + getSessionId() (default = getQueryId): a connector reads the credential and the stable AuthSession key (preserved across FE observer->master forwarding) off the session. - ConnectorCapability.SUPPORTS_USER_SESSION: gates FE credential injection (least-privilege -- a connector that never consumes the token never receives it). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_013KKXx3btbQc6w3MbZbEoRf
…injection + per-user cache bypass FE bridge for the Iceberg REST per-user session feature (builds on the SPI DTO): - ConnectorSessionBuilder.from(ctx): copy the retained per-connection DelegatedCredential + sessionId onto the ConnectorSession ONLY when the target connector declares SUPPORTS_USER_SESSION (least-privilege); map fe-core DelegatedCredential -> neutral ConnectorDelegatedCredential by enum name so an added-but-unmapped type fails loud rather than being silently dropped. - PluginDrivenExternalCatalog.shouldBypassTableNameCache / shouldBypassDbNameCache = supportsUserSession() && ctx.hasDelegatedCredential(). Under session=user the REST server returns PER-USER metadata, so the shared (catalog+name-keyed, NOT user-keyed) db/table-name caches are bypassed to prevent cross-user leakage; a session with no credential keeps the shared cache (the fail-closed rejection then happens connector-side on the actual read, never by serving/poisoning a shared entry). - ExternalCatalog/ExternalDatabase: db-level live-path bypass (getFilteredDatabaseNames(false)/getDbNullableWithoutCache/matchesLocalDbName), still re-appending information_schema+mysql and never mutating the shared lowerCaseTo*Name lookup when bypassing (updateLookup=false). Tests: ConnectorSessionImplTest (capability-gated inject / non-capable skip / no-cred), PluginDrivenExternalCatalogSessionBypassTest (the bypass decision across capability + credential + null-context). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_013KKXx3btbQc6w3MbZbEoRf
…outing (metadata + scan/write/procedure) Iceberg connector consumer for the per-user REST session feature: - IcebergConnectorProperties + IcebergRestMetaStoreProperties: iceberg.rest.session (none|user) + iceberg.rest.oauth2.delegated-token-mode (access_token|token_exchange) + optional session-timeout, with validation (session=user requires security.type=oauth2; relax the "oauth2 requires static token/credential" rule for user-session, whose identity is the per-request user token). - IcebergDelegatedCredentialUtils + IcebergSessionCatalogAdapter: bridge the neutral ConnectorDelegatedCredential to an iceberg SessionCatalog.SessionContext (access_token = verbatim OAuth2 bearer; token_exchange = the typed token-type key) over a SINGLE shared RESTSessionCatalog. delegatedCatalog/delegatedViewCatalog fail closed on a tokenless session=user request (never borrow a shared identity). - IcebergConnector: build a RESTSessionCatalog for a session=user REST catalog; newCatalogBackedOps(session) routes per-request metadata through the querying user's delegated catalog; declare SUPPORTS_USER_SESSION only for that config. - scan/write/procedure providers take a Function<ConnectorSession, IcebergCatalogOps> resolver (connector passes this::newCatalogBackedOps); the legacy ops constructors bind a constant s->ops so existing tests/behaviour stay byte-identical. This makes SELECT / INSERT-DELETE-MERGE / ALTER TABLE ... EXECUTE run per-user. The write COMMIT was already per-user (IcebergConnectorTransaction is opened by beginTransaction over the session-aware metadata ops). Tests: IcebergProviderSessionRoutingTest (each provider resolves ops with the call session + fail-closed), IcebergSessionCatalogAdapterTest (credential-key mapping per mode + fail-closed + fromString), IcebergConnectorValidatePropertiesTest (+session validation). IcebergConnectorTest/IcebergProcedureOpsTest adjusted for the resolver constructor (reflection reads catalogOpsResolver; a bare-null ctor arg is cast). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_013KKXx3btbQc6w3MbZbEoRf
…research notes + HANDOFF (Tasks 0-7 DONE) - Add the authoritative design doc (Trino-aligned, SPI-retargeted) and the source-verified research notes (Trino reference + apache#63068 as-built + current SPI). - HANDOFF: mark Tasks 0-7 DONE with the 3-commit stack (SPI 55c991d / FE 930e477 / connector a07216e), the clean-room review result (routing completeness = zero gaps; cache-leakage = none), and the two follow-ups (background log-noise reduction; two minor case/consistency nits). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_013KKXx3btbQc6w3MbZbEoRf
…essionContextTest (cross-user cache-leakage data-flow) Closes the one remaining apache#63068 test gap: the DATA-FLOW proof (not just the bypass decision, which PluginDrivenExternalCatalogSessionBypassTest already pins) that a session=user catalog serves per-user database metadata LIVE and never through the shared (catalog+name-keyed, not user-keyed) name cache -- the cross-user leakage guard (Trino CVE-2026-34214). Retargeted off the deleted IcebergRestExternalCatalog onto a PluginDrivenExternalCatalog test subclass: mockStatic(SessionContext.current()) drives the per-token identity, the catalog overrides the remote listing to return each user's own databases and record the listing token. Two disjoint per-user results + a live re-list on EVERY read (including a repeat of an earlier token) prove no user's database set leaks to another and nothing was served from a shared cache; information_schema+mysql stay visible. apache#63068 asserted this via a no-token "bootstrap" read, which on this branch fail-closes (a session=user catalog has no shared identity to bootstrap) -- the per-read live token record is the architecture-correct equivalent observable. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_013KKXx3btbQc6w3MbZbEoRf
…a-flow test + Task 0 retained-base verify Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_013KKXx3btbQc6w3MbZbEoRf
…er rebase to 4f48ccf Rebase branch-catalog-spi onto upstream-apache/master (10 upstream commits, 21 local commits replayed). Two files were touched by both sides; both are paimon + apache#65365 fallout: 1. Legacy fe-core PaimonScanNode.java: modify/delete conflict — upstream apache#65365 added "jni.enable_file_reader_async" to its BACKEND_PAIMON_OPTIONS forwarding list, while P5 (T29) deleted the whole legacy fe-core paimon subsystem. Resolution keeps the deletion and ports the new key to the connector-side mirror list PaimonScanPlanProvider.BACKEND_PAIMON_JNI_OPTIONS (+ intent test): the SPI path is the only forwarding path left, so without the port the new catalog property would silently never reach BE's PaimonJniScanner. 2. PaimonJniScanner.java: silent semantic conflict — apache#65365 removed the then-unused java.util.Collections import upstream, while this branch's null-predicate backstop uses Collections.emptyList(). git auto-merged the non-overlapping hunks without complaint, breaking compilation. Restore the import. Verified: fe-connector-paimon package GREEN (PaimonScanPlanProviderTest 65/0 incl. new backendOptionsForwardFileReaderAsyncOptOut); paimon-scanner Paimon* tests 11/0; full build.sh --fe BUILD SUCCESS (build cache disabled). Pre-existing local-env JVM crash in java-common JniScannerTest (StubRoutines::jbyte_disjoint_arraycopy_avx3, byte-identical module before and after rebase) is unrelated and left as-is. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
…hade Under a parallel reactor (build.sh's default -T 1C), the package-phase maven-shade-plugin goal can start before this no-source module's main jar is created and attached, failing with "Failed to create shaded artifact, project main artifact does not exist". build.sh's single-threaded retry regex does not match this error string, so it surfaces as BUILD FAILURE instead of auto-retrying. Pin maven-jar-plugin's default-jar execution to the package phase with forceCreation=true, declared before the shade plugin, so the module's main artifact is guaranteed present before shade runs -- eliminating the race while keeping -T parallelism (no fallback to single-threaded). Verified: bash build.sh --fe --clean (default -T 1C, all 61 modules) BUILD SUCCESS, the race error did not recur, jar:jar (default-jar) runs before shade:shade, and the shaded-jar relocation output is unchanged (the diff is jar-plugin-only). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
…on_deletes connector gap as flip blocker Rebase: 23 upstream commits / 23 ours replayed, behind0/ahead23. All 5 conflicts came from a single upstream commit 0814e49 (apache#65135, iceberg $position_deletes system table), which lands in fe-core code P6 deleted/migrated to the connector SPI. Records the per-file adjudication, the two hazards git did not flag as conflicts (IcebergScanNodeTest hard-depends on a method the P6 side removes; the groovy suite's auto-merged out-of-hunk lines), and the resulting flip blocker: the iceberg connector does not expose position_deletes, so the feature is absent vs upstream master while its ~1200 lines of BE readers land orphaned. Two p0 suites are intentionally left red rather than editing the tests to hide the gap. Verified: FE_MAVEN_THREADS=1 -Dmaven.build.cache.enabled=false bash build.sh --fe = BUILD SUCCESS, 61 modules, 0 checkstyle violations. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
…+ design (T0 evidence in, ready to implement) Research (workflow wf_583cf18a-7ae, 5 lenses) + T0 empirical prerequisites (workflow wf_2f84859d-a52, 3 checks that actually compiled and ran code). T0 falsified two design assumptions, both now recorded: - Jackson vs Gson: Float/Double are byte-identical (the suspected risk did not reproduce); BigDecimal differs in BOTH directions (1.50 -> 1.5, and 10 -> 1E+1, a syntax change). Root cause is JsonNodeFactory bigDecimalExact=false under valueToTree, not the writer. Fix: a hoisted mapper copy with withExactBigDecimals(true) -- never mutate iceberg's shared JsonUtil.mapper(). - partition_data_json is NOT parsed as JSON: it is fed to Doris's STRUCT text serde, and DataTypeNullableSerDe::from_string swallows every parse error into a NULL partition while returning OK. A wrong shape is silent wrong data, not an error. The connector's existing renderer emits a JSON array and would hit this. Also found a blocker-shaped gap: [D-065] skips the schema-evolution dict for all sys handles on the rationale that the metadata-table schema rides inside the serialized FileScanTask -- true only for the JNI path. position_deletes is the first sys table reaching a native BE reader, which needs history_schema_info to resolve the `row` column (v1 hard-errors, v2 silently degrades to name matching). Design decisions: D1 smooth-upgrade guard NOT ported (user ruling: implement the final form, no compat) => fe-core and SPI both unchanged, the port is entirely connector-local. D2 third range shape on IcebergScanRange. D3/D5/D6 per T0. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
… design (T0 evidence in) Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
…ble on the connector SPI Ports upstream apache#65135 (0814e49) onto this branch's connector SPI. Upstream implemented $position_deletes in the fe-core iceberg subsystem that P6 deleted, so after the rebase the BE readers, the thrift contract and the two regression suites were present while no FE ever produced the ranges -- the feature was absent vs master and its ~1200 lines of BE C++ were orphaned. fe-core and the SPI are UNCHANGED: the smooth-upgrade backend guard is deliberately not ported (user ruling: implement the final form), which was the only piece needing engine-side state. Everything lands inside fe-connector-iceberg. - IcebergConnectorMetadata: expose position_deletes (drop the two Q2-era filters). The exclusion mirrored legacy's "not supported yet"; apache#65135 invalidated that premise. - IcebergScanRange: a THIRD range shape. BE routes into its native reader only on top-level TIcebergFileDesc.content in {1,3} + a PARQUET/ORC range format, and asserts exactly one delete descriptor. The data path must therefore never set top-level content to 1 or 3 -- now a load-bearing invariant. - IcebergPartitionUtils: getPartitionDataObjectJson, a SECOND renderer. The existing one emits a JSON array of strings; this path needs a type-native object keyed by the metadata table's partition field NAMES resolved by iceberg field ID. Values ride a mapper copy with withExactBigDecimals(true): stock Jackson renders BigDecimal("10") as 1E+1, a JSON syntax change vs legacy Gson (verified empirically). - IcebergScanPlanProvider: the planning branch (newBatchScan -- PositionDeletesTable .newScan() throws -- pin, predicate, split sizing, PUFFIN-vs-parquet/orc, AVRO fails loud), and a narrowed [D-065] gate: every other sys table's schema rides inside the serialized FileScanTask, but this one reads natively and BE needs history_schema_info to resolve `row` (v1 hard-errors, v2 silently degrades to name matching). Two defects found by adversarial review and fixed here: the dict branch called resolveSysTable, which carries no auth wrap by contract and would have failed a kerberized catalog at plan time (now built from the already-loaded base table, also saving a round-trip); and the shared split-size heuristic summed fileSizeInBytes, which over-counts puffin DVs ~N-fold (now ScanTaskUtil.contentSizeInBytes, byte-equal for data files -- the comment claiming iceberg 1.10.1 lacks that method was false). Verified: connector suite 977/977 (+11 new), 0 checkstyle violations, and FE_MAVEN_THREADS=1 -Dmaven.build.cache.enabled=false bash build.sh --fe = BUILD SUCCESS, 61 modules. The two p0 regression suites are NOT run -- no local iceberg REST+MinIO+Spark env; they remain the flip gate. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
…remaining flip gate Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Refs apache#65185. Part of **P7 — hive (+HMS)** in the Catalog SPI migration tracking issue. This PR starts the hive/HMS connector migration on `branch-catalog-spi`, covering the HMS metadata path, metastore-event pipeline, ACID transaction write path, and the hudi live cutover dependency tracked in P7. --------- Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
… a slim HMS metastore-client shade (apache#65721) ## Proposed changes Part of the catalog SPI migration umbrella: apache#65185. The `fe/fe-connector` closure depended on `org.apache.doris:hive-catalog-shade`, a ~122 MB "kitchen-sink" shade whose Hive metastore-client content is under 2% of its bytes (the rest is a full Hadoop, the paimon bundle, an ancient fastutil, DLF, etc.). This PR replaces it, for the fe-connector modules, with a new plugin-private slim shade `fe-connector-hms-hive-shade` (~15 MB) that bundles **only** the Hive metastore-**client** closure and relocates `org.apache.thrift` to `shade.doris.hive.org.apache.thrift`, mirroring the existing `fe-connector-paimon-hive-shade`. ### What changed - **New module `fe-connector-hms-hive-shade`**: hive-standalone-metastore (api + client), hive-common (`HiveConf`), hive-serde, hive-storage-api, iceberg-hive-metastore (`HiveCatalog`), libthrift/libfb303 0.9.3, codehaus jackson 1.9.2 (HMS JSON event messaging) and commons-lang 2.x (`HiveConf`). It uses an artifactSet `<includes>` **whitelist** so the hadoop-yarn/curator/jersey/kerby tail dragged in by hive-shims stays out; hadoop-common and guava/slf4j/log4j come from each plugin's own deps / the host at runtime. The thrift relocation **reuses** the existing `shade.doris.hive` prefix, so the vendored patch `HiveMetaStoreClient` in `fe-connector-hms` needs no source change. - **`fe-connector-hms`** now depends on the slim shade (compile, transitive to hive/hudi/iceberg) instead of the fat shade; the patch client stays in place. - **`fe-connector-iceberg`** drops its direct fat-shade dependency in the **same commit** (the hms-flavor `HiveCatalog` now arrives transitively via the slim shade), so iceberg is never left with two `HiveCatalog` copies; adds `iceberg-bundled-guava` at compile scope for the vendored `DeleteFileIndex`. - **`fe/pom.xml`** keeps the `hive-catalog-shade` version pin (still used by `be-java-extensions`) with an explanatory comment; stale comments in the iceberg/hudi poms are corrected. ### Size impact The shade artifact drops **122 MB → 15 MB** (to ~12% of its size). Each of the hive / hudi / iceberg plugins bundles its own copy, so each deployed plugin shrinks by **~107 MB** (~322 MB across the three). ### Verification - `fe-connector-hms` / `-hive` / `-hudi` / `-iceberg` build + unit tests green (197 test classes, 0 failures, checkstyle 0). - `dependency:tree` across all 19 fe-connector modules shows no `hive-catalog-shade`. - The three assembled plugin zips contain the slim shade once, no fat shade, no original-package libthrift, and exactly one copy each of the metastore-api `Table` / `HiveConf` / iceberg `HiveCatalog` / relocated thrift; the metastore and `iceberg.hive` bytecode is md5-identical to the fat shade. - **Remaining gate**: heterogeneous-HMS e2e (hive read/write, iceberg-on-HMS INSERT/DELETE/MERGE, hudi-on-HMS read; TCCL / filter-hook / kerberos paths) — to be exercised by CI / a follow-up. 🤖 Generated with [Claude Code](https://claude.com/claude-code) https://claude.ai/code/session_01AkwFY4c7kZkLKZeazy3UMW --------- Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
…er updates; next = PERF-06 Records the vended-catalog table-comment cache (PERF-05, commit aea3ebd): design doc (with the re-scoping vs PERF-01 and the red-team security fix that narrows the gate to vended-non-session), summary, tasklist check-off, progress log, and a fresh HANDOFF pointing at PERF-06 (the per-file vended StorageProperties rebuild, C3) with the fe-core DefaultConnectorContext routing decision flagged. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01SMtYwYyyubZZiC1odLZTG3
…g once per scan, not per file (PERF-06) Root cause: for a REST catalog with vended credentials, iceberg scan planning re-derived the whole StorageProperties map (StorageProperties.createAll + a hadoop Configuration build) for EVERY data file and EVERY delete file, via context.normalizeStorageUri(uri, token). But the vended token is scan-invariant (extracted once, the same Map instance threads through every range builder), and the derivation is a pure function of the token, independent of the per-file URI. So a 50k-file scan burned tens of seconds of FE CPU (doubled on merge-on-read) rebuilding an identical map O(N_files + N_deletes) times. Fix (connector-side "prepare once, apply many"): add a scan-scoped SPI seam ConnectorContext.newStorageUriNormalizer(token) that returns a UnaryOperator<String> which bakes the token-derived config in once and then normalizes each URI cheaply. DefaultConnectorContext overrides it to build the effective map lazily-memoized on the first non-empty URI, byte-identical to normalizeStorageUri(uri, token) (same empty-uri short-circuit, same vended-replaces-static precedence, same fail-loud LocationPath, same exception timing). The SPI default folds back to a per-call normalizeStorageUri, so every other connector is unaffected. The iceberg provider builds the normalizer once at each of the three scan-start sites (synchronous data, streaming, position_deletes) and threads it through buildRange / convertDelete / buildPositionDeleteRange in place of the token; TcclPinningConnectorContext delegates the new method to the raw context so iceberg gets the hoist (this path is pure fe-core, no TCCL pin needed). DefaultConnectorContext is per-catalog and shared across queries, so this scan-scoped hoist deliberately keeps no cross-query state and retains no credentials (the normalizer is a scan-local, single-threaded, GC'd at scan end). Not in scope: the write path getBackendFileType is O(1) per write, not per data file. Measurement / guard: one vended scan now derives the config once instead of O(N_files + N_deletes) times. A new connector-side guard test asserts a 3-file scan enters newStorageUriNormalizer exactly once while normalizing all three paths; new fe-core parity tests assert the normalizer's output is byte-identical to normalizeStorageUri across the vended / static / empty-uri / fail-loud cases. Tests: fe-connector-iceberg full module 974 UT green (0 fail / 1 skip); fe-core DefaultConnectorContextNormalizeUriTest 13 UT green. Checkstyle clean. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01SMtYwYyyubZZiC1odLZTG3
…er updates; next = PERF-07 Record the route-A decision (connector-side "prepare once, apply many" via a new scan-scoped SPI normalizer seam, rejecting the fe-core per-catalog memo for its cross-query thrash + credential-retention risk), the adversarial red-team outcome (lazy-memoized effective for exact exception-timing parity), and the two-stage build verification (fe-core must be built separately since the iceberg connector does not depend on it). Roll the tasklist/progress/HANDOFF forward to PERF-07 (C20 write-path 3-5 loads of the same table). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01SMtYwYyyubZZiC1odLZTG3
…(PERF-07) Introduce ConnectorStatementScope, a neutral SPI memoization arena that lives exactly as long as one SQL statement, so a connector can load a table (and derive per-statement state) once and share that one object across every read + write resolver in the statement. - fe-connector-api: ConnectorStatementScope (interface + NONE no-op) and a default ConnectorSession.getStatementScope() returning NONE, so all existing session implementations are unaffected. - fe-core: ConnectorStatementScopeImpl (ConcurrentHashMap-backed); a lazily built StatementContext field with a synchronized accessor plus a reset (not cleared on close/release, reclaimed with the statement, mirroring snapshots); ConnectorSessionImpl/Builder capture the scope reference at construction (the request thread, where the statement context is reachable) so off-thread scan pumps that reuse the one session still reach it; and ExecuteCommand resets the scope on each prepared-statement execution so a reused context starts fresh. This is the fe-core landing point for Trino's per-transaction metadata-cache model, which Doris cannot host on the ~26-times-rebuilt connector session but can hang on the one per-statement StatementContext. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01SMtYwYyyubZZiC1odLZTG3
…ent, drop the fat handle and the delete stash (PERF-07) Route the read metadata path, scan planning, write shaping and beginWrite through the per-statement ConnectorStatementScope so one statement loads each iceberg table once and every resolver shares that one RAW object; the snapshot pin and the Kerberos doAs FileIO wrap are applied per consumer, never frozen into the shared object. - New IcebergStatementScope keys the shared table by catalogId:db:table:queryId and the rewritable-delete supply by catalogId:queryId; a null/NONE session loads every time (byte-identical to the pre-scope offline behavior). - Remove the fat-handle memo (IcebergTableHandle.resolvedTable + accessors + the three with* carries); the handle is now pure coordinates. - beginWrite resolves the shared table instead of loading fresh (a row-level DML reuses the object the scan already loaded); openTransaction keeps newTransaction()'s refresh, giving the commit a fresh OCC base. - Move the merge-on-read rewritable-delete supply out of the per-catalog IcebergRewritableDeleteStash singleton (deleted) into the scope: the scan seam accumulates, the write seam drains, both by the same key. A format-version>=3 DELETE/UPDATE/MERGE under a NONE scope now fails loud instead of silently writing a deletion vector that would drop old deletes and resurrect rows. Read, scan and write of one statement resolve ops through the same per-session newCatalogBackedOps(session), so sharing one table stays within one user's identity (session=user / REST-vended safe). There is no write-side latency goal; the value is architectural coherence and removing the singleton + class. iceberg 968 pass / 1 skip; fe-core scope + session tests green. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01SMtYwYyyubZZiC1odLZTG3
… updates; next = PERF-08 Record the completed PERF-07 (unified per-statement table-load owner): summary doc, progress entry (6-way recon + adversarial review notes + build gotchas), tasklist row flipped to done, and a fresh HANDOFF pointing at PERF-08 (the rewrite_data_files / expire_snapshots maintenance-path re-scan/dedup fixes). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01SMtYwYyyubZZiC1odLZTG3
…sign docs Commit the two design docs the tracker/summary reference: the unified per-statement table-owner design (authoritative) and the superseded narrow resolution-owner design it replaces. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01SMtYwYyyubZZiC1odLZTG3
…ration into one union scan (PERF-08 / C19) Root cause: ConnectorRewriteDriver STEP3 registered each bin-packed rewrite group's source files with a separate connectorTx.registerRewriteSourceFiles call. The iceberg connector re-derives those raw paths back to DataFiles with a full-table planFiles() scan on every call, all against the SAME once-pinned OCC snapshot (startingSnapshotId, captured at beginWrite). So G groups triggered G+1 full-table manifest scans; at G ~ 50-200 that is minutes of pure planning IO. Fix: union every group's paths into one Set and register once (unionSourceFilePaths). registerRewriteSourceFiles already accepts a Set<String> and accumulates across calls, and every per-group call scanned the identical pinned snapshot, so one union scan reconstructs exactly the same filesToDelete set: bin-packed groups are path-disjoint and the connector dedups by path, so no double-registration; the four result counters come from group metadata, not filesToDelete; any unresolvable path still fails loud + rolls back in both variants. G planFiles() scans collapse to 1. No change to result row, committed state, or success/failure outcome. Connector side unchanged. Constraint: fe-core edit only converges an existing driver's calls to an existing neutral SPI (Set<String>, no source-specific branch, no logic moved into fe-core) -- within the "fe-core source only shrinks / stays connector-agnostic" rules. Gate: the distributed STEP3 needs a live cluster (deferred to the flip rehearsal e2e), so unionSourceFilePaths is extracted and unit-tested for union/cross-group dedup/empty-group/empty-plan parity. ConnectorRewriteDriverTest 5/5 green; 0 checkstyle violations. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01SMtYwYyyubZZiC1odLZTG3
…anifest reads across snapshots (PERF-08 / C21) Root cause: IcebergExpireSnapshotsAction.buildDeleteFileContentMap built the delete-file -> content classification map by looping over every snapshot and reading all of that snapshot's delete manifests, with no visited-path dedup. Iceberg manifests are immutable and adjacent snapshots carry the same delete manifests forward unchanged, so a single manifest is re-read once per referencing snapshot -- S snapshots x M manifests of serial remote reads, largely redundant. Fix: add a method-scoped visited set and skip a delete manifest whose path was already read (if (!visitedDeleteManifests.add(manifest.path())) continue). Reading each DISTINCT manifest exactly once collapses the reads from O(S*M) to O(distinct M). The result map is byte-identical: an immutable manifest yields the same DeleteFile set on every read, and putIfAbsent first-writer-wins is order- independent because a delete file's content type is immutable. visited MUST stay at method scope (a comment guards this) -- inside the snapshot loop it would reset each iteration and defeat the cross-snapshot dedup. Gate: new @VisibleForTesting lastDeleteManifestReadCount records the distinct manifests read; buildDeleteFileContentMap made package-visible. New test seeds a table whose delete manifest is carried across multiple snapshots (position + equality deletes), asserts reads == distinct manifests (strictly fewer than the per-snapshot total) and that both delete files stay classified as POSITION_DELETES / EQUALITY_DELETES. IcebergExpireSnapshotsActionTest 9/9 green; 0 checkstyle violations. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01SMtYwYyyubZZiC1odLZTG3
…er updates; next = P2 remaining Records the maintenance-path fixes (rewrite_data_files union registration C19, expire_snapshots delete-manifest dedup C21): the design + red-team verdict, the summary with both commit hashes, tasklist marked done, a progress entry, and the HANDOFF rewritten to point at the remaining P2 tasks (PERF-09/10/11). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01SMtYwYyyubZZiC1odLZTG3
…ariants across a file's byte-slices (PERF-11 / C12+C15a+C13) Root cause: TableScanUtil.splitFiles cuts one data file into k byte-slice FileScanTasks, and buildRange ran per slice, recomputing values that are identical for the whole file every time: partition_data_json (date/time format + Jackson), the ordered identity partitionValues, and the merge-on-read delete carriers. On a partitioned MOR table at 100k splits that is ~0.5-2s of extra planning CPU plus large garbage; the k ranges also each held their own copies (FE heap), and a v3 scan re-put an identical rewritable-delete supply per slice. Fix: thread a 1-entry per-file cache (PerFileScratch) through buildRangeForTask, the shared choke point of the eager planScanInternal loop and the streaming IcebergStreamingSplitSource. computePerFileInvariants runs once per file (keyed by the shared DataFile instance, which all k slices return from task.file()); the file's remaining slices reuse the cached values, and the k ranges share the same immutable partitionValues Map and deleteCarriers List. Only start/length and the size-proportional selfSplitWeight stay per-slice. The v3 rewritableDeleteSupply put is moved to once per file, on the file's FIRST slice keyed to the new file (covering the last file in the stream — a flush-on-evict would drop it and silently resurrect deleted rows). The streaming scratch is a per-source (= per-scan) field, single-threaded, so it stays O(1) memory (streaming OOM safety preserved). count-pushdown passes a null scratch (single range, no caching). Byte-identical: every cached value is a deterministic function of task.file() and scan-invariant inputs; distinct logical files never share a DataFile instance (the SDK and manifest-cache enumeration both copy per entry), so the identity key never returns a stale value; splitFiles emits a file's slices consecutively, so the 1-entry cache hits 100% (a hypothetical non-consecutive arrival would only cost perf, never correctness); IcebergScanRange is immutable and every consumer reads only. Verified by a 3-lens adversarial byte-parity review (0 breaks). Out of scope (separate follow-ups): the wire-side per-range TFileRangeDesc payload duplication needs a thrift delete-file dictionary + BE reader change (protocol evolution, C15b); the fe-core generic-node per-split hoists are a framework-layer change (C14). Gate: @VisibleForTesting perFileInvariantComputeCount + two tests (a split file computes its invariants once, not per slice; two split files compute twice with no cross-file staleness). IcebergScanPlanProviderTest 105/105, IcebergConnectorTransactionTest 66/66, full iceberg module 1064 pass / 1 skip / 0 fail; 0 checkstyle violations. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01SMtYwYyyubZZiC1odLZTG3
…ral: design/summary/tracker PERF-11 done partially (C12 per-file invariant memo + C15a instance sharing + C13 plan-side put-once, commit 10b7d29): design + 3-lens byte-parity red-team verdict + summary; C15b (wire-bytes thrift+BE protocol change) and C14 (fe-core generic-node hoists) recorded as deferred with their prerequisites. PERF-10 (WHERE conjunct conversion) re-verified as low-value and not cleanly fixable within the no-framework-touch constraint; marked deferred with the revival condition. Tasklist rows, progress log, and HANDOFF (next = PERF-09 or PERF-11 remainder) updated. 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01SMtYwYyyubZZiC1odLZTG3
…er handle (PERF-11 / C14) Root cause: PluginDrivenScanNode.resolveScanProvider() = connector.getScanPlanProvider(currentHandle), and the SPI contract builds providers fresh/stateless per call. It is invoked on the per-split / per-range hot path (getFileCompressType per split, getDeleteFiles per range, plus ~10 planning sites), so every split re-allocated a scan provider + did two TCCL classloader swaps to run what is an identity no-op for iceberg. All three enumeration paths hit it (eager coordinator loop, concurrent partition-batch appendBatch threads, single streaming pump). Lowest-magnitude tier (~100-200ns/split, no remote IO), but pure waste: the provider is a function of currentHandle, which is only refined during early pushdown/pin, before split enumeration. Fix: memoize the resolved provider keyed on currentHandle IDENTITY, held in an immutable (handle, provider) holder published via a single volatile write. A hit reuses the instance; an identity miss (first call, or currentHandle swapped by pushdown/pin) re-resolves and republishes. A null provider (no scan capability) is cached and returned correctly. Because it is keyed on the exact currentHandle each caller would pass, it is byte-identical to calling connector.getScanPlanProvider(currentHandle) every time - no reliance on any connector's provider-selection semantics. The final fields + single volatile write make it safe for the concurrent partition-batch appendBatch threads (no torn new-key/old-provider read); sharing one provider instance across the scan is already the established pattern on the heavier planScan path (startSplit/startStreamingSplit capture and share one provider across concurrent async tasks). Connector-agnostic (no source branching); connectors that field-cache their provider (jdbc/maxcompute) see a strict no-op, connectors that new-per-call (iceberg/paimon/hive) lose the per-split allocation - no connector regresses. Measurement / gate: PluginDrivenScanNodeScanProviderSelectionTest. memoizesProviderForStableHandleAndReResolvesOnHandleChange asserts getScanPlanProvider is invoked exactly once for a stable handle across repeated resolves (was N-per-split; observed red "Wanted 1 time ... but was 3" before the fix) and re-resolves once on a handle change. Full PluginDrivenScanNode*Test family: 103 pass / 0 fail / 0 skip; fe-core Checkstyle 0 violations. fe-core-only change; the iceberg connector does not depend on fe-core and is unaffected. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01SMtYwYyyubZZiC1odLZTG3
…P2-remainder recon; tracker/handoff updates - Design + summary for the resolveScanProvider() per-handle memo (fix committed in 87ff73b). - Records the session-4 recon of the three remaining P2 candidates (all evidence re-verified live, not stale): the streaming-pump micro-batch (fe-core framework, needs sign-off because micro-batching changes the backend split distribution - not byte-identical), the C15b/C13 wire-byte dedup (protocol evolution: thrift delete-file dictionary + BE reader + mixed-version compat, needs sign-off), and the C14 generic-node slices #1/#2/#4 (deferred: cross-connector Hadoop-Path byte risk / new SPI capability surface). User chose only C14 slice #3 (provider memo). - tasklist/HANDOFF: mark C14 provider-memo done; HANDOFF next-step = the two sign-off-gated blocks only. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01SMtYwYyyubZZiC1odLZTG3
…ing spaces - plan-doc/master-todo/: a cross-task register of deferred, decision-gated big items (needs user sign-off or protocol evolution), so they are not forgotten. Seeds two entries with full background/current-call-stack/solution-call-stack/example writeups: the streaming split-dispatch micro-batch (fe-core; not byte-identical -> changes backend split distribution) and the wire-side delete-file dictionary (thrift + BE reader + mixed-version compat). - plan-doc/per-statement-table-owner-port/: tracking space for porting iceberg's PERF-07 per-statement table-load owner pattern to other connectors. Key finding recorded: the neutral fe-core/fe-connector-api scaffolding (ConnectorStatementScope + ConnectorSession.getStatementScope + StatementContext plumbing) is already in place, so porting is connector-side-only work. Candidate map (to be scoped next session): paimon (high) / hive-hms (mid-high) / hudi (mid); maxcompute/es/jdbc/trino likely excluded (no per-statement metastore loadTable fan-out). No product code touched. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01SMtYwYyyubZZiC1odLZTG3
morningman
requested review from
924060929,
englefly,
morrySnow and
starocean999
as code owners
July 19, 2026 01:18
Contributor
|
Thank you for your contribution to Apache Doris. Please clearly describe your PR:
|
morningman
force-pushed
the
branch-catalog-spi
branch
from
July 19, 2026 02:50
b2dff68 to
7b3821f
Compare
morningman
requested review from
CalvinKirs,
Gabriel39,
dataroaring,
gavinchou,
liaoxin01,
luwei16,
mymeiyi and
seawinde
as code owners
July 19, 2026 02:50
morningman
force-pushed
the
branch-catalog-spi
branch
from
July 20, 2026 04:31
5f00959 to
3fb9cae
Compare
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
No description provided.