-
Notifications
You must be signed in to change notification settings - Fork 3.9k
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
bazel: reported stress failure but test is passing #74013
Comments
This test did not fail (If I'm reading the logs correctly). I saw this in another failure recently too (can't find it now, but I did assign that issue to the bazel folks as well so it's in your proj board). |
From the logs:
|
Unfortunately the |
I think I found the same in the other issue. Fixing that would be a good first step. edit: spent a good five minutes looking for that other issue I keep referring to. No dice. Not sure what I did with it but I can't imagine I did anything but rename it + file it for dev-inf and bazel, like this one. |
Found it: #73841 |
We've seen some issues where Bazel jobs are failing in `github-post` (cockroachdb#73841, cockroachdb#74013) with the following output: found outstanding output. Considering last test failed: It's hard to say what the problem is because these scripts haven't kept the `test.json.txt` in `artifacts`. Here I remove the logic to clean up the file so we can RC further instances of the problem. Release note: None
OK, PR to fix that problem: #74152 |
72992: util/log: fix redactability of logging tags r=abarganier a=knz Fixes #72905. Some time in the v21.2 cycle, the log entry preparation logic was refactored and a mistake was introduced: the logging tags were not any more subject to the redaction logic. The result was that redaction markers were missing in log tag values, and if a value had contained unbalanced redaction markers in a value string (say, as part of a SQL table key), it would have caused log file corruption and possibly a confidential data leak. This patch fixes that, by preparing the logging tags in the same way as the main message for each entry. Release note (cli change): A bug affecting the redactability of logging tags in output log entries has been fixed. This bug had been introduced in the v21.2 release. 73937: setting: introduce setting classes r=RaduBerinde a=RaduBerinde This commit introduces the three setting classes in the RFC (#73349): `SystemOnly`, `TenantReadOnly`, and `TenantWritable`. The `SystemOnly` class replaces the existing `WithSystemOnly()`. In this change we don't yet implement the advertised semantics. We mechanically use `TenantWritable` for all settings except those that were using `WithSystemOnly()` which use `SystemOnly`; this should not change any existing behavior. The classes will be revisited in a separate change, after we implement the semantics. Release note: None 73978: opt: fix like escape processing for span constraints r=cucaroach a=cucaroach Fixes: #44123 Previously no attempt was made to properly handle escape ('\\') sequence in like patterns being turned into constraints. Refactor code used to process like at runtime to generate a regexp and use that to properly handle index constraint generation. Release note (sql change): Escape character processing was missing from constraint span generation which resulted in incorrect results when doing escaped like lookups. 74102: sql: do not fetch virtual columns during backfill r=mgartner a=mgartner Fixes #73372 Release note (bug fix): A bug has been fixed that caused internal errors when altering the primary key of a table. The bug was only present if the table had a partial index with a predicate that referenced a virtual computed column. This bug was present since virtual computed columns were added in version 21.1.0. 74110: bazel: require setting `cockroach_cross=y` to opt into cross toolchains r=irfansharif a=rickystewart With `--incompatible_enable_cc_toolchain_resolution` set in #73819, now Bazel selects the appropriate toolchain for you. Bazel was selecting the `cross_linux_toolchain` when building for the host platform on Linux, resulting in link errors when trying to compile `stress` under `race`. We update the toolchains to instead require opting into the cross toolchains by defining `cockroach_cross=y`. Closes #73997. Release note: None 74111: bench/rttanalysis: allow roundtrips to be off by 1 r=ajwerner a=ajwerner If we don't have a range, let the currently estimate be wrong by 1. We mostly care about the ballpark and the growth rate. I'm sick of these flakes. Fixes #73884. Release note: None 74152: ci: don't delete `test.json.txt` after processing r=tbg a=rickystewart We've seen some issues where Bazel jobs are failing in `github-post` (#73841, #74013) with the following output: found outstanding output. Considering last test failed: It's hard to say what the problem is because these scripts haven't kept the `test.json.txt` in `artifacts`. Here I remove the logic to clean up the file so we can RC further instances of the problem. Release note: None Co-authored-by: Raphael 'kena' Poss <[email protected]> Co-authored-by: Radu Berinde <[email protected]> Co-authored-by: Tommy Reilly <[email protected]> Co-authored-by: Marcus Gartner <[email protected]> Co-authored-by: Ricky Stewart <[email protected]> Co-authored-by: Andrew Werner <[email protected]>
We've seen some issues where Bazel jobs are failing in `github-post` (cockroachdb#73841, cockroachdb#74013) with the following output: found outstanding output. Considering last test failed: It's hard to say what the problem is because these scripts haven't kept the `test.json.txt` in `artifacts`. Here I remove the logic to clean up the file so we can RC further instances of the problem. Release note: None
@rickystewart we have multiple reproductions with the output now, the output always ends mid-stream, for example here: Could dev-inf pick this up again? I'm seeing a few flakes per week from this and I imagine there's a fair amount that don't get routed to me. |
More fallout from this cc @mgartner https://cockroachlabs.slack.com/archives/C023S0V4YEB/p1641931553012600 |
Maybe as a hot fix, we can disable the nightly bazel stress build? |
Done. |
I've preserved the incomplete output from one of the "failing" tests here. |
@mgartner -- Broken link. |
@rickystewart Oops. Fixed. |
|
Since #75087, this should be resolved. |
Unfortunately, this doesn't seem to be resolved, see #76492. |
If I look in the build log I see the following:
|
I'm going to re-close this because it doesn't seem to have anything to do with the original bug (which, indeed, I have not seen repro'd since I fixed it a few weeks back). |
@rickystewart should we file an issue to make stray panics easily discoverable in bazel test logs/CI? |
@irfansharif I think so. We have #74105 and it seems related, but not exactly the same as, the panic logging issue. Let me know if you don't file the issue and I will. |
@tbg, I followed the sequence of steps above starting from the TC failure link. I feel in this specific instance the panic logging was not that undiscoverable. Probably there are other instances where there isn't the case, but if they come up, I'll file separately with evidence where we can improve. Scrolling below: |
@irfansharif my pre-bazel muscle memory has me ignore the build log. All failures were detectable from artifacts. In particular, one could look at full_output.txt, which is where I looked, and where the output ends abruptly: I do think if it is straightforward to arrange, panics should be included there as well (this was certainly the old behavior). It's problematic if there are multiple places one has to look for any given kind of failure. |
I agree it's problematic if you have to look in one of multiple places to find failures, and for that reason I think that dropping the pre-Bazel muscle memory and checking the build log is a good idea. There are some exceptions -- Regarding getting panic info into the
@rail wrote this, I think. Looks like we convert the |
It's not in the json file (https://teamcity.cockroachdb.com/repository/download/Cockroach_Nightlies_StressBazel/4369269:id/full_output.tgz!/kvserver_test.test.json.txt) as far as I can tell. |
ts/testmodel.TestModelDBQuery failed with artifacts on master @ 3f95a4bd83cce2952a12497de82692866b4da659:
Help
See also: [How To Investigate a Go Test Failure \(internal\)](https://cockroachlabs.atlassian.net/l/c/HgfXfJgM)Parameters in this failure:
This test on roachdash | Improve this report!
The text was updated successfully, but these errors were encountered: