panic crash while on the database pages #83935
Assignees
Labels
C-bug
Code not up to spec/doc, specs & docs deemed correct. Solution expected to change code/behavior.
T-sql-queries
SQL Queries Team
Comments
maryliag
added
the
C-bug
|
I was testing again, this time without using -insecure, and it switches between the panic above and another one Video of reproducing the panic: https://www.loom.com/share/9a6f143ae950448b851ffd860ddb05d7 |
|
The branch I'm using is from this PR #83677 (if it helps). I did a rebase and rebuilt, and then got those panics |
|
Thanks, I can repro on your branch. |
|
I'm somewhat confident that #83615 is to blame - it somehow exposed some issue with releasing of the prepared statements in some cases, still looking. |
|
Alright, I have figured out the cause of the first stack trace and made some progress on the second. I believe it occurs when I still don't fully understand how we get those "leftover bytes" errors - I'm pretty sure it has to do with the memory account used in In short, a couple of small PRs (which I'm about to open) plus #83678 should address this. |
craig bot
pushed a commit
that referenced
this issue
Jul 8, 2022
83597: Colocate auth logging with auth metric for consistency r=rafiss a=ecwall refs #83224 Release note (bug fix): Move connection OK log and metric to same location after auth completes for consistency. This resolves an inconsistency (see linked isssue) in the DB console where the log and metric did not match. 83731: kvserver: acquire replica lease on queue check r=nvanbenschoten a=kvoli This patch adds a check within the replication for when a replica is the raft leader and does not have a valid lease. The necessary conditions are that it is currently the raft leader and that the lease status is expired. This ensures that following a node restart, a replicas with a valid lease will be installed within the replica scanner interval. **single nodes 10k ranges with change**  **5 node, 3k ranges** *without change*  *with change*  resolves #83444 Release note: None 84044: ui: option to search exact statement on SQL Activity r=maryliag a=maryliag Previously, when doing a search on SQL Activity page, it was returning all statements that contained all terms from the search, but not necessarily on the same order. This commit adds an option when you wrap the search in quotes it will only return results with the exact match in order. https://www.loom.com/share/442c6eaee84b4c71a1acdef0b63b74bf Release note (ui change): Ability to search for the exact terms in order when wrapping the search in quotes. 84047: sql: remove unused error return value in a method of connExecutor r=yuzefovich a=yuzefovich Found while looking into #83935. Release note: None 84082: roachtest: skip multitenant/fairness r=cucaroach a=cucaroach Informs: #83994 Release note: None 84085: roachtest: fix zipping of artifacts to include other zips r=srosenberg a=renatolabs When artifacts are zipped in preparation for being published to TeamCity, other zip files are skipped. The idea is that we won't try to recursively zip artifacts.zip itself, or debug.zip, which is published separately. However, some tests (notably, `tpchvec`) download their own zip files in the `logs` directory so that they'll be available for analysis when a test fails. While there was an intention to skip only top-level zip files (as indicated by existing comments), the code itself would skip any zip files found in the artifacts directory. This commit updates the zipping logic to skip only toplevel zip files, allowing tests to write their own zip files to the `logs` directory and have them available for inspection later. Release note: None. Co-authored-by: Evan Wall <[email protected]> Co-authored-by: Austen McClernon <[email protected]> Co-authored-by: Marylia Gutierrez <[email protected]> Co-authored-by: Yahor Yuzefovich <[email protected]> Co-authored-by: Tommy Reilly <[email protected]> Co-authored-by: Renato Costa <[email protected]>
craig bot
pushed a commit
that referenced
this issue
Jul 9, 2022
84048: row: only store the accounted for memory if the reservation is approved r=yuzefovich a=yuzefovich Previously, we would update the counter about the reserved memory before doing the reservation. If that reservation is denied, then later on, in `txnKVFetcher.close` we could try to release more memory than we registered. This is now fixed. Addresses: #83935. Release note: None Co-authored-by: Yahor Yuzefovich <[email protected]>
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Seeing a panic crash, (other people mentioned they saw on other occasions, but these are the steps I was able to reproduce):
Created a build with
make buildStarted CRDB with
./cockroach demo --insecure --multitenant=falseStart the console with
make ui-watch TARGET=http://localhost:8080/Open the db console on one of the database list of tables, e.g. http://localhost:3000/#/database/system
Wait for awhile (sometime it took me a few minutes, sometimes 30min, sometimes doesn't happen at all) and it will crash and the trace shows on the terminal
Trace:
Jira issue: CRDB-17362
The text was updated successfully, but these errors were encountered: