bazel/dev: add fix/warning for long TMPDIR's, especially on problem OS's (Mac/BSD) #72918
Labels
A-build-system
B-os-macos
Issues specific to macOS.
C-enhancement
Solution expected to add code/behavior + preserve backward-compat (pg compat issues are exception)
T-dev-inf
Comments
rickystewart
added
C-enhancement
This was referenced May 31, 2022
rickystewart
added a commit
to rickystewart/cockroach
that referenced
this issue
Jun 6, 2022
Bazel's default behavior of rooting the `tmpdir` to an "in-sandbox" directory has been a point of confusion for CRL developers. The sandbox directory does not exist after the test is run (unless `--sandbox_debug` is provided), which is sometimes confusing for folks who expect their test's temp files to be present where the logs suggest they should be (see cockroachdb#82413). Furthermore, the long `tmpdir` used in these cases breaks tests that create Unix sockets on OS's where Unix sockets have a maximum path length. Avoid these problems by having `doctor` just tell you to manually set a `test_tmpdir`. We add `/tmp` to `gitignore` in case people want to root it at the `tmp` directory in their checkout. Closes cockroachdb#72918. Closes cockroachdb#82413. Release note: None
craig bot
pushed a commit
that referenced
this issue
Jun 7, 2022
81782: sql: use declarative schemachange for add column sequence exprs r=fqazi a=fqazi Fixes: #81781 Previously, the declarative schema changer was disabled for add column expressions with sequence references (i.e. default, on update, computed) because we were missing telemetry from the legacy schema changer for during backfill related failures. This was inadequate because the lack of telemetry from these areas could be seen as a usability issue. To address this, this patch adds in support for missing telemetry and enables supports for add column operations with sequence operations. Release note: None 82388: sql: remove date/intervalstyle_enabled from code r=otan,mgartner a=rafiss fixes #81529 No release note is needed, since v22.1 already included a release note about how these are both hardcoded to true and cannot be changed. The session variables and cluster settings remain in the code, but are marked as retired. Release note: None 82451: kvserver/rangefeed: fix off-by-one in `NewCatchUpIterator()` r=miretskiy a=erikgrinaker **kvserver/rangefeed: fix off-by-one in NewCatchUpIterator()** `NewCatchUpIterator` created the `MVCCIncrementalIterator` with `RangeFeedRequest.Header.Timestamp.Prev()`, because it assumed the given timestamp was inclusive. However, the rest of the rangefeed code treats this timestamp as exclusive. This patch uses the correct, exclusive timestamp when creating the `MVCCIncrementalIterator`. This change has no externally visible effect: even though the previous code would cause `MVCCIncrementalIterator` to emit keys at `Timestamp`, the `CatchUpScan()` method would discard those events if they were at or below `Timestamp`. An integration test is also added to verify that the start timestamp of a rangefeed catchup scan is exclusive. The test passes both with the new and the old code, as expected. Release note: None **rangefeed: emphasize that start time is exclusive** This patch adds comments for rangefeed-related APIs and components emphasizing that the start timestamp of rangefeeds is exclusive. In other words, the first possible emitted event (including catchup scans) will be at `RangeFeedRequest.Header.Timestamp.Next()`. Several parameters are also renamed to emphasize this. There are no functional or semantic changes. Touches #82488. Release note: None 82466: dev: have `doctor` advise to set a particular `tmpdir` r=rail a=rickystewart Bazel's default behavior of rooting the `tmpdir` to an "in-sandbox" directory has been a point of confusion for CRL developers. The sandbox directory does not exist after the test is run (unless `--sandbox_debug` is provided), which is sometimes confusing for folks who expect their test's temp files to be present where the logs suggest they should be (see #82413). Furthermore, the long `tmpdir` used in these cases breaks tests that create Unix sockets on OS's where Unix sockets have a maximum path length. Avoid these problems by having `doctor` just tell you to manually set a `test_tmpdir`. We add `/tmp` to `gitignore` in case people want to root it at the `tmp` directory in their checkout. Closes #72918. Closes #82413. Release note: None 82476: update cluster-ui to v22.2.0-prerelease-1 r=maryliag a=maryliag Update cluster-ui to latest published version Release note: None 82519: stats: fix flaky test TestDefaultColumns r=rytaft a=rytaft TestDefaultColumns creates statistics on a table with 110 columns using the command `CREATE STATISTICS s FROM t.a`. It then checks that there are exactly 101 column statistics on table t.a with statistics_name = 's' (one stat for the primary index, plus 100 other column stats). However, this test may be flaky if automatic statistics are running, since each new automatic stat will cause other stats to be deleted. Although the test disables automatic stats at the beginning, it seems that some sort of race condition may cause it to be reenabled. This commit fixes the problem by disabling automatic stats using the new table level settings, ensuring that the 101 column stats are not deleted after they have been created. Fixes #81513 Release note: None Co-authored-by: Faizan Qazi <[email protected]> Co-authored-by: Rafi Shamim <[email protected]> Co-authored-by: Erik Grinaker <[email protected]> Co-authored-by: Ricky Stewart <[email protected]> Co-authored-by: Marylia Gutierrez <[email protected]> Co-authored-by: Rebecca Taft <[email protected]>
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
This was first spotted in #72027. The TMPDIR that Bazel gives tests is in the sandbox and is therefore quite long, which can introduce problems especially on macOS OS's. Tests can therefore fail in surprising ways. A couple ideas:
--test_tmpdirto something appropriate. So on macOS we could check indev doctorwhether you have--test_tmpdirset in your~/.bazelrc/.bazelrc.user. This is problematic though if you have a habit of setting--test_tmpdiron the command-line, since the two flags will conflict.rules_goto check whether you're running on macOS and whetherTMPDIRis too long and overrideTMPDIRaccordingly. We actually already setTMPDIRin our fork ofrules_go, so the logic would fit right in there.I think (2) makes more sense, but the blanket override could be surprising.
Epic CRDB-8036
Jira issue: CRDB-11345
The text was updated successfully, but these errors were encountered: