Conversation
… op and return the old one
…s with threadPool for threaded worker add_disk operations
…, added function to parse message properly and drain message bytes properly
…e partitions on the disks to be replaced to the list
…t immediately if check_all_disks failed
|
I'll just post here what we went over on Slack as to include information for others or in case we need it
Other than my comment for the magic number and what I've outlined above, this LGTM |
|
Well just because it's non blocking doesn't mean it won't soak up a CPU core. I've made that mistake before and it helps to insert a small sleep to let the CPU do something else |
|
@cholcombe973 I initially brought up the same thing, but reluctantly decided we may not need to ensure that resource gets dunked on. That being said, weird stuff happens, and that weird stuff tends to be the stuff that bites back. @mzhong1 I think we should probably completely promise that thread won't get blocked in the loop |
…non-lvm OSDs and fix database ticket update function to use correct column name
…at operations once finished running
…ent exploding log sizes
|
just a suggestion at this point but I think this should either be merged or split up. It's quite large |
Yups, I figure at this point I will probably squash merge this if everyone is alright with this PR.... |
|
Go for it |
…n the disk instead of the partition only
4 participants
Completely overhaul the Socket system to use STREAM sockets instead of REQ/REP sockets, adding in multithreading support for the disk-manager, and message queue handling for bynar and disk-manager.
This is in an effort to support disk-manager operations that may take more than a few minutes to execute, allowing Bynar to run operations without repeating and without hanging on a single operation for possibly hours depending on the operation.