[WIP] PartitionStoreManager exclusive ownership by Worker role #3606
Conversation
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
This was referenced Jul 30, 2025
This was referenced Jul 31, 2025
This introduces a few crucial changes to how we handle panics in restate. Prior to this change, we would abort the process at panic time without considering a clean shutdown nor rocksdb wal fsync. The summary of changes is as follows: - We now always unwind the stack on panics. TaskCenter is designed to catch panics of important tasks and trigger a clean shutdown and reports a non-zero exit code. - Ensure that on graceful shutdown timeout that we attempt to cleanly flush/shutdown rocksdb manager. This is important to avoid massive backfills of lost memtables on unclean shutdown. - Catch panics at top-level task-center runtime control loop and trigger an emergency rocksdb WAL fsync to ensure that we flush the WAL to avoid loss of in-memory WAL buffer if/when we add support to manual wal flushing in the future. - Makes sure that panics from network connection tasks do not trigger a system shutdown, instead, they are caught and properly logged. This avoids a situation where a network bad request/handler can cause the entire node to panic. - In situations where tracing might have been lost/dropped, ensure that we also log critical information on stderr. This hardens restate against unclean crashes and ensures we perform a clean handoff to other cluster members in case of an unrecoverable crash.
This allows the worker to flush rocksdb as soon as the worker role is stopped and before the rest of the system is shut down. In order to achieve this, datafusion queries will always use the remote scanner even on worker nodes. This adds serialization/memory cost that would have been otherwise avoided but potentially can be optimized in future PRs.
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
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.
This allows the worker to flush rocksdb as soon as the worker role is stopped and before the rest of the system is shut down. In order to achieve this, datafusion queries will always use the remote scanner even on worker nodes. This adds serialization/memory cost that would have been otherwise avoided but potentially can be optimized in future PRs.
Stack created with Sapling. Best reviewed with ReviewStack.