-
Notifications
You must be signed in to change notification settings - Fork 1.9k
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
Fix assert by only accessing idx #6924
Conversation
Asserting on `_rowCount < Utils.Size(_valueBoundaries)` was catching a case where `_rowCount`'s update was reordered before `_valueBoundaries` This was unnecessary, since this method doesn't need to use `_rowCount`. Instead, make the asserts use only `idx` which will be maintained consistent with the waiter logic in this cache. �Ensure we only ever use `_rowCount` from the caching thread, so write reordering won't matter.
I tested this fix in #6922 as well |
Codecov ReportAttention:
Additional details and impacted files@@ Coverage Diff @@
## main #6924 +/- ##
=======================================
Coverage 68.79% 68.79%
=======================================
Files 1249 1249
Lines 249431 249433 +2
Branches 25510 25509 -1
=======================================
+ Hits 171604 171607 +3
- Misses 71214 71216 +2
+ Partials 6613 6610 -3
Flags with carried forward coverage won't be shown. Click here to find out more.
|
This seems to have fixed things, but now our failures are all timeouts. Hitting here and in #6923. @michaelgsharp didn't you mention wanting to address the timeouts by splitting up the test assemblies? |
@ericstj Yeah, as soon as I finish the NER issue I'll test splitting the assemblies, so I'll have a test out for that later today. |
Asserting on `_rowCount < Utils.Size(_valueBoundaries)` was catching a case where `_rowCount`'s update was reordered before `_valueBoundaries` This was unnecessary, since this method doesn't need to use `_rowCount`. Instead, make the asserts use only `idx` which will be maintained consistent with the waiter logic in this cache. �Ensure we only ever use `_rowCount` from the caching thread, so write reordering won't matter.
* Update dependencies from https://github.com/dotnet/arcade build 20231220.2 Microsoft.DotNet.Arcade.Sdk , Microsoft.DotNet.Build.Tasks.Feed , Microsoft.DotNet.Helix.Sdk , Microsoft.DotNet.SignTool , Microsoft.DotNet.SwaggerGenerator.MSBuild , Microsoft.DotNet.XUnitExtensions From Version 8.0.0-beta.23265.1 -> To Version 8.0.0-beta.23620.2 * Fixed version update breaks. * Update XUnitVersion * Update MicrosoftMLOnnxRuntimeVersion to 1.16.3 * Rollback OnnxRuntime and suppress warning * Update to Xunit with fix for xunit/xunit#2821 * Update Centos docker containers * Fix packaging step * Try including stdint.h to fix missing uint8_t on centos * Update Centos test queue * Attempt to use runtime centos-stream8-helix container for tests * Use centos-stream8-mlnet-helix container for testing * Undo changes to test data * Make NETFRAMEWORK ifdef versionless * Only use semi-colons for NoWarn * Fix assert by only accessing idx (#6924) Asserting on `_rowCount < Utils.Size(_valueBoundaries)` was catching a case where `_rowCount`'s update was reordered before `_valueBoundaries` This was unnecessary, since this method doesn't need to use `_rowCount`. Instead, make the asserts use only `idx` which will be maintained consistent with the waiter logic in this cache. �Ensure we only ever use `_rowCount` from the caching thread, so write reordering won't matter. * Don't include the SDK in our helix payload (#6918) * Don't include the SDK in our helix payload I noticed that the tests included the latest SDK - including the host - in our helix payloads. This is a large amount of unnecessary downloads and it also makes it so we use the latest host on the older frameworks which can fail when the latest host drops support for distros. Since our tests shouldn't need the full CLI, remove this from our helix payloads. We'll instead get just the runtime we need through `AdditionalDotNetPackage` * Place Helix downloaded runtime on the PATH Helix only sets the path when the CLI is included, however we don't need the CLI. * Make double assertions compare with tolerance instead of precision (#6923) Precision might cause small differences to round to a different number. Instead compare with a tolerance which is not sensitive to rounding. --------- Co-authored-by: dotnet-maestro[bot] <dotnet-maestro[bot]@users.noreply.github.com> Co-authored-by: Michael Sharp <misharp@microsoft.com> Co-authored-by: Eric StJohn <ericstj@microsoft.com>
Fixes #6921
Asserting on
_rowCount < Utils.Size(_valueBoundaries)
was catching a case where_rowCount
's update was reordered before_valueBoundaries
This was unnecessary, since this method doesn't need to use
_rowCount
.Instead, make the asserts use only
idx
which will be maintained consistent with the waiter logic in this cache. This will lock and synchronize with the main thread to ensureidx
is always within bounds.Ensure we only ever use
_rowCount
from the caching thread, so write reordering won't matter.