Repository navigation
server: log Substrate actors that have no Task record at startup - #510
Open
itssimrank wants to merge 4 commits into
Open
itssimrank wants to merge 4 commits into
itssimrank wants to merge 4 commits into
Conversation
After a Redis restart, Substrate actors can keep running with no Task record. On startup, list all actors, page through all tasks, and log any actor with no matching task. Log only; nothing is deleted. The check runs before the HTTP server starts, with a 30 second timeout. If Substrate is not configured, it is skipped.
Paging tasks by offset could skip a task when one was deleted between pages, which made the check log a live actor as an orphan. A single read of the task index avoids that. ListTasks skips index entries with no record, so a short result no longer means the end of the list.
Substrate v0.5.0 moved page size and page token into ListOptions on ListActorsRequest. Bump the pin to main's version and read them there.
itssimrank
changed the title
mayawang
approved these changes
Oct 24, 2026
mayawang
left a comment
mayawang
left a comment
Collaborator
There was a problem hiding this comment.
Thanks @rkb32 for the rebase and for testing against a live v0.5.0 cluster, and @itssimrank for opening this and catching the license headers. Tests pass on main. Approving.
Member
|
We shouldn't merge this, querying all actors is a very expensive query when there are millions of actors in a deployment. We shouldn't put an expensive query at the startup time. Instead, we need to persist the Task metadata persistently. We are going to make some major changes to the storage layer to avoid shadow copies of actors. Please let's not merge this. |
Member
|
See #515. |
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
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.
A Redis restart drops every Task record while the Substrate actors keep running, so ax no longer knows about sandboxes that still hold worker slots (#442). Make the divergence visible: at startup, list all actors and all tasks and log a warning for each actor with no matching atespace/name. The check only logs, is bounded to 30s, and a failure does not stop the server.
Tasks are read in one call rather than paged by offset, because a delete between pages would shift the list and could skip a live task.
Carried in from rkb32/ax:orphan-check at the maintainer's request on the issue; implementation and tests are rkb32's.
Part of #442