Condition Agent nodes hang 15+ seconds on tiktoken.pages.dev DNS failure - need graceful fallback

[Mordris]

Describe the bug

When using OpenAI models (gpt-4o-mini, gpt-3.5-turbo, gpt-4, etc.) in Condition Agent nodes, Flowise attempts to fetch tiktoken tokenizer data from tiktoken.pages.dev for pre-calculation of token counts. When this domain is unreachable due to network restrictions (DNS issues, firewalls, corporate proxies), the node hangs for 15+ seconds with multiple retry attempts before falling back to approximate counting.

This issue is specific to Condition Agent nodes only - regular Agent nodes, QnA nodes, and other specialist agents do not exhibit this behavior because they don't require pre-execution token counting.

Error logs:

flowise | Failed to calculate number of tokens, falling back to approximate count TypeError: fetch failed
flowise | at node:internal/deps/undici/undici:13510:13
flowise | at process.processTicksAndRejections (node:internal/process/task_queues:95:5)
flowise | at async RetryOperation._fn (/usr/local/lib/node_modules/flowise/node_modules/p-retry/index.js:50:12) {
flowise | [cause]: Error: getaddrinfo EAI_AGAIN tiktoken.pages.dev
flowise | at GetAddrInfoReqWrap.onlookupall [as oncomplete] (node:dns:122:26) {
flowise | errno: -3001,
flowise | code: 'EAI_AGAIN',
flowise | syscall: 'getaddrinfo',
flowise | hostname: 'tiktoken.pages.dev'
flowise | }
flowise | }

Impact:

• Response times increase from 2-3 seconds to 15+ seconds
• Production deployments unusable in corporate/restricted networks
• Forced to use non-OpenAI models despite licensing/requirements

To Reproduce

1. Deploy Flowise in an environment where tiktoken.pages.dev is unreachable (corporate network, firewall blocking Cloudflare Pages, DNS restrictions)
2. Create an agentflow with a Condition Agent node
3. Configure the Condition Agent to use any OpenAI model (gpt-4o-mini, gpt-3.5-turbo, gpt-4)
4. Send a message through the agentflow
5. Observe logs showing repeated DNS failures to tiktoken.pages.dev
6. Notice 15+ second delay before response is generated

Network diagnostics showing the issue:

# DNS resolution fails
$ docker compose exec flowise nslookup tiktoken.pages.dev
;; connection timed out; no servers could be reached

# But other domains work fine
$ docker compose exec flowise nslookup google.com
Name:   google.com
Address: 142.251.140.46

Important: The issue does NOT occur with:

• Google Gemini models (gemini-pro, gemini-2.5-flash) in Condition Agent
• OpenAI models in regular Agent nodes (non-condition)
• Other specialist agents (QnA, Database Query, etc.)

Expected behavior

Condition Agent nodes should handle tiktoken fetch failures gracefully:

1. Fast timeout: Fail fast (1-2 seconds max) instead of hanging for 15+ seconds
2. Immediate fallback: Switch to approximate token counting without blocking
3. Configuration option: Provide environment variable to disable tiktoken fetching entirely:
   • DISABLE_TIKTOKEN=true
   • USE_APPROXIMATE_TOKENS=true
   • TIKTOKEN_TIMEOUT=1000 (milliseconds)
4. Graceful degradation: Log warning but continue execution immediately

The node should function normally with approximate token counting when tiktoken is unavailable, similar to how it eventually behaves after the timeout.

Screenshots

flowise     | Failed to calculate number of tokens, falling back to approximate count TypeError: fetch failed
flowise     |     at node:internal/deps/undici/undici:13510:13
flowise     |     at process.processTicksAndRejections (node:internal/process/task_queues:95:5)
flowise     |     at async RetryOperation._fn (/usr/local/lib/node_modules/flowise/node_modules/p-retry/index.js:50:12) {
flowise     |   [cause]: Error: getaddrinfo EAI_AGAIN tiktoken.pages.dev
flowise     |       at GetAddrInfoReqWrap.onlookupall [as oncomplete] (node:dns:122:26) {
flowise     |     errno: -3001,
flowise     |     code: 'EAI_AGAIN',
flowise     |     syscall: 'getaddrinfo',
flowise     |     hostname: 'tiktoken.pages.dev'
flowise     |   }
flowise     | }
flowise     | Failed to calculate number of tokens, falling back to approximate count TypeError: fetch failed
flowise     |     at node:internal/deps/undici/undici:13510:13
flowise     |     at process.processTicksAndRejections (node:internal/process/task_queues:95:5)
flowise     |     at async RetryOperation._fn (/usr/local/lib/node_modules/flowise/node_modules/p-retry/index.js:50:12) {
flowise     |   [cause]: Error: getaddrinfo EAI_AGAIN tiktoken.pages.dev
flowise     |       at GetAddrInfoReqWrap.onlookupall [as oncomplete] (node:dns:122:26) {
flowise     |     errno: -3001,
flowise     |     code: 'EAI_AGAIN',
flowise     |     syscall: 'getaddrinfo',
flowise     |     hostname: 'tiktoken.pages.dev'
flowise     |   }
flowise     | }

Flow

Issue test Agents.json

Use Method

Docker

Flowise Version

3.0.11

Operating System

None

Browser

None

Additional context

No response