Replace the shard manager's pod-keyed routing table with an executor-keyed ShardLeaseState - #3757
Conversation
✅ Deploy Preview for golemcloud canceled.
|
| } | ||
| } | ||
|
|
||
| // TODO(shard manager redesign, ticket 2): remove together with the CAS persistence migration |
There was a problem hiding this comment.
let's not commit such references to ongoing tickets in comments
|
The backward compatibility is unnecessary - you can remove it to make the codebase simpler. The Golem 1.6 release is going to be breaking in many ways. |
…keyed ShardLeaseState
… old rows via migration
9169fdd to
e2b2e89
Compare
Done. |
|
CI note: the group6 failure is oplog_processor_shard_reassignment_no_loss (5 != 6 at sharding.rs:1218). I reproduced it locally on pristine main (748d197) as well as on this branch — ~1 in 3 attempts on both, masked by #[flaky(3)] most of the time. It's the "no checkpoint" re-delivery the test's own comment describes, not related to the routing-table change. worker-tests-group1's failure (long_running_poll_loop_interrupting_and_resuming_by_second_invocation) passed locally on this branch too. Re-running the workflow. |
2 participants
1/7 of the ShardManager Redesign.
Resolves GOL-445