tetragon: Teach file writer to exit gracefully #185
Merged
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
The context 'cancel()' done on closing does not propagate into the ProcessManager and Server objects, which currently use the basic context.Background(). The problem with this is on exit we hook SIGTERM, etc. and call cancel() but, because this doesn't get pushed back to the JSON writer object its possible to have partial writes completed. First to clean this lets ensure we use the ctx from main so cancel will be pushed into these objects. Signed-off-by: John Fastabend <[email protected]>
When we receive a signal -- sigterm for example -- we tear down the BPF state and event handlers in go routines with cancel(). But, we do not wait to ensure they have received the signal and done any cleanup needed. This adds a waitgroup to use for this. So objects increment it on New() create op and then call the Done() when they are complete. Ensuring we don't exit in the middle of sensitive operations, such as writes to gRPC streams and JSON files where we can generate corrupted events when we are unlucky. Signed-off-by: John Fastabend <[email protected]>
jrfastab
force-pushed
the
pr/jrfastab/ctxprop
branch
from
7b8a132 to
7b4ea4d
Compare
kkourt
reviewed
Jun 23, 2022
jrfastab
force-pushed
the
pr/jrfastab/ctxprop
branch
from
7b4ea4d to
4b7c8d1
Compare
kkourt
approved these changes
Jun 24, 2022
willfindlay
approved these changes
Jun 24, 2022
michi-covalent
approved these changes
Jun 24, 2022
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
We observed that rarely if we SIGTERM Tetragon (e.g. what pod restart and destroy do before SIGKILL it) that a corrupted json file might get left behind. The theory is this is do to writers being killed in the middle of an operation.
To fix we do two things. First propagate a single context through the processManager and Server objects instead of just using context.Backgroun(). This allows the cancel() call from our SIGTERM handler to be handled by ctx.Done() in the respective go routines. Then we add a waitGroup to the signal handler to ensure it waits until the writer has exited.
I only fix the go routine responsible for writes here. We should audit the other routines as well.
And second I started thinking about how we nest the various objects, at the moment we pass a lot of input types through the New() constructors. But this looked bigger than a fix so held off for now on this.