structured logging: replacing Fatal/Exit/etc. without loss of flushing #303
Conversation
|
/assign @thockin @serathius |
| "k8s.io/klog/v2" | ||
| ) | ||
|
|
||
| func ExampleFlushAndExit() { |
There was a problem hiding this comment.
Any reason why we can't have a TestFlushAndExit (Similar to functions in TestErrorS in klog_test.go) ?
We can check against the following format.
format: "\"exiting...\"\nos.Exit(10)"
There was a problem hiding this comment.
We could have such a test function, but an example is more useful:
- it gets tested by
go test - it gets included in the documentation
|
/assign @dims For approval + final LGTM. |
| // Applications that replace those calls and do not have some specific | ||
| // requirements like "exit immediately" can use this value as parameter | ||
| // for FlushAndExit. | ||
| ExitFlushTimeout = 10 * time.Second |
There was a problem hiding this comment.
are end users able to modify these directly? they shouldn't right?
There was a problem hiding this comment.
I don't see a bit need for modifying it, but also no good reason why we shouldn't allow it. Being able to modify them is mostly (just?) useful in tests.
This timeout can become a constant.
The OsExit function has to be a variable, but we could unexport it.
There was a problem hiding this comment.
Speaking of OsExit, if we allow modifying it, should Fatal and friends also call it?
There is a feature request where someone is asking for Fatal to not call os.Exit. We cannot change the default behavior and not call os.Exit (that behavior is part of the stable API), but if we export OsExit and use it in Fatal, then someone who knows his application can perhaps change it into a function which panics.
There was a problem hiding this comment.
I've pushed a version that uses OsExit everywhere instead of calling os.Exit directly and also updated the comments: https://github.com/kubernetes/klog/compare/dbc5aa051e765a4216e0fb89e0b54fae92ad8ab0..56b8a4ea35ba0fc3be9fc82182e71c2f0116adc1
When structured logging called for functions like Fatal and Exit to be avoided in favor of klog.ErrorS + os.Exit, it was overlooked that this replacement skips flushing. We could ask developers to use klog.ErrorS + klog.Flush + os.Exit, but that is still not quite the same because klog.Flush could get stuck or take too long. Therefore klog now supports flushing + exiting with a dedicated function that can be combined with some other, arbitrary logging function. Allowing applications to change the default timeout and how specifically the application terminates may be useful, so both values are variables that are exported and users of klog are allowed to change.
|
/lgtm /hold |
|
/hold cancel /approve |
|
[APPROVALNOTIFIER] This PR is APPROVED This pull-request has been approved by: dims, pohly, serathius The full list of commands accepted by this bot can be found here. The pull request process is described here
Needs approval from an approver in each of these files:
Approvers can indicate their approval by writing |
What this PR does / why we need it:
When structured logging called for functions like Fatal and Exit to be avoided
in favor of klog.ErrorS + os.Exit, it was overlooked that this replacement
skips flushing.
We could ask developers to use klog.ErrorS + klog.Flush + os.Exit, but that is
still not quite the same because klog.Flush could get stuck or take too long.
Therefore klog now supports flushing + exiting with a dedicated function that
can be combined with some other, arbitrary logging function.
Which issue(s) this PR fixes (optional, in
fixes #<issue number>(, fixes #<issue_number>, ...)format, will close the issue(s) when PR gets merged):Fixes #301
Special notes for your reviewer:
See also https://kubernetes.slack.com/archives/C020CCMUEAX/p1644341190882399
Please confirm that if this PR changes any image versions, then that's the sole change this PR makes.