flaky usdt test #2402
Comments
|
I'm fighting with it in 4eb280b but I can't make it stable yet. Now the pidof invocation to get the child pid after it has started is failing. I may just make that logic only run if the test needs BEFORE_PID. |
|
I think that this should be resolved by #2370 |
|
Unfortunately, I don't think it was fully fixed, just made a bit better. I still don't fully understand how this happens (and why it happens on that test specifically) but here's a run after that PR that timed out again - https://github.com/BurntBrunch/bpftrace/actions/runs/3381817256/jobs/5616467787 |
BurntBrunch
pushed a commit
to BurntBrunch/bpftrace
that referenced
this issue
Nov 3, 2022
As noted in bpftrace#2402, usdt flakiness was made better by 508538a but not fully fixed. This commit is what I should have done all along: it allows the test runner to parse and wait for multiple BEFORE clauses and thus ensures the processes have started before the test runs. There are two minor changes: 1. The check for child processes is now `ps --ppid` based to eventually allow parallel process runs in the same environment. 2. Because of the `ps` usage, the name check is now truncated to 15 chars, which will fail if TASK_COMM_LEN is not 16. That looks like a constant in the kernel, so I think we're good.
BurntBrunch
pushed a commit
to BurntBrunch/bpftrace
that referenced
this issue
Nov 3, 2022
As noted in bpftrace#2402, usdt flakiness was made better by 508538a but not fully fixed. This commit is what I should have done all along: it allows the test runner to parse and wait for multiple BEFORE clauses and thus ensures the processes have started before the test runs. There are two minor changes: 1. The check for child processes is now `ps --ppid` based to eventually allow parallel process runs in the same environment. 2. Because of the `ps` usage, the name check is now truncated to 15 chars, which will fail if TASK_COMM_LEN is not 16. That looks like a constant in the kernel, so I think we're good.
BurntBrunch
pushed a commit
to BurntBrunch/bpftrace
that referenced
this issue
Nov 3, 2022
As noted in bpftrace#2402, usdt flakiness was made better by 508538a but not fully fixed. This commit is what I should have done all along: it allows the test runner to parse and wait for multiple BEFORE clauses and thus ensures the processes have started before the test runs. There are two minor changes: 1. The check for child processes is now `ps --ppid` based to eventually allow parallel process runs in the same environment. 2. Because of the `ps` usage, the name check is now truncated to 15 chars, which will fail if TASK_COMM_LEN is not 16. That looks like a constant in the kernel, so I think we're good.
BurntBrunch
pushed a commit
to BurntBrunch/bpftrace
that referenced
this issue
Nov 4, 2022
As noted in bpftrace#2402, usdt flakiness was made better by 508538a but not fully fixed. This commit is what I should have done all along: it allows the test runner to parse and wait for multiple BEFORE clauses and thus ensures the processes have started before the test runs. There are two minor changes: 1. The check for child processes is now `ps --ppid` based to eventually allow parallel process runs in the same environment. 2. Because of the `ps` usage, the name check is now truncated to 15 chars, which will fail if TASK_COMM_LEN is not 16. That looks like a constant in the kernel, so I think we're good.
|
Okay, I give up. #2414 doesn't help either, so this is not about the test runner racing the BEFORE calls. Maybe it's something to do with --usdt-file-activation and whatever kernel the GH runners run? |
viktormalik
pushed a commit
to viktormalik/bpftrace
that referenced
this issue
Apr 20, 2023
As noted in bpftrace#2402, usdt flakiness was made better by 508538a but not fully fixed. This commit is what I should have done all along: it allows the test runner to parse and wait for multiple BEFORE clauses and thus ensures the processes have started before the test runs. There are two minor changes: 1. The check for child processes is now `ps --ppid` based to eventually allow parallel process runs in the same environment. 2. Because of the `ps` usage, the name check is now truncated to 15 chars, which will fail if TASK_COMM_LEN is not 16. That looks like a constant in the kernel, so I think we're good.
viktormalik
pushed a commit
to viktormalik/bpftrace
that referenced
this issue
Apr 20, 2023
As noted in bpftrace#2402, usdt flakiness was made better by 508538a but not fully fixed. This commit is what I should have done all along: it allows the test runner to parse and wait for multiple BEFORE clauses and thus ensures the processes have started before the test runs. There are two minor changes: 1. The check for child processes is now `ps --ppid` based to eventually allow parallel process runs in the same environment. 2. Because of the `ps` usage, the name check is now truncated to 15 chars, which will fail if TASK_COMM_LEN is not 16. That looks like a constant in the kernel, so I think we're good.
viktormalik
pushed a commit
to viktormalik/bpftrace
that referenced
this issue
Apr 20, 2023
As noted in bpftrace#2402, usdt flakiness was made better by 508538a but not fully fixed. This commit is what I should have done all along: it allows the test runner to parse and wait for multiple BEFORE clauses and thus ensures the processes have started before the test runs. There are two minor changes: 1. The check for child processes is now `ps --ppid` based to eventually allow parallel process runs in the same environment. 2. Because of the `ps` usage, the name check is now truncated to 15 chars, which will fail if TASK_COMM_LEN is not 16. That looks like a constant in the kernel, so I think we're good.
viktormalik
pushed a commit
to viktormalik/bpftrace
that referenced
this issue
Apr 20, 2023
As noted in bpftrace#2402, usdt flakiness was made better by 508538a but not fully fixed. This commit is what I should have done all along: it allows the test runner to parse and wait for multiple BEFORE clauses and thus ensures the processes have started before the test runs. There are two minor changes: 1. The check for child processes is now `ps --ppid` based to eventually allow parallel process runs in the same environment. 2. Because of the `ps` usage, the name check is now truncated to 15 chars, which will fail if TASK_COMM_LEN is not 16. That looks like a constant in the kernel, so I think we're good.
viktormalik
pushed a commit
to viktormalik/bpftrace
that referenced
this issue
Apr 20, 2023
As noted in bpftrace#2402, usdt flakiness was made better by 508538a but not fully fixed. This commit is what I should have done all along: it allows the test runner to parse and wait for multiple BEFORE clauses and thus ensures the processes have started before the test runs. There are two minor changes: 1. The check for child processes is now `ps --ppid` based to eventually allow parallel process runs in the same environment. 2. Because of the `ps` usage, the name check is now truncated to 15 chars, which will fail if TASK_COMM_LEN is not 16. That looks like a constant in the kernel, so I think we're good.
viktormalik
pushed a commit
to viktormalik/bpftrace
that referenced
this issue
Apr 20, 2023
As noted in bpftrace#2402, usdt flakiness was made better by 508538a but not fully fixed. This commit is what I should have done all along: it allows the test runner to parse and wait for multiple BEFORE clauses and thus ensures the processes have started before the test runs. There are two minor changes: 1. The check for child processes is now `ps --ppid` based to eventually allow parallel process runs in the same environment. That requires to use `ps` from the `procps` package on Alpine as the default BusyBox one doesn't have the `--ppid` option. 2. Because of the `ps` usage, the name check is now truncated to 15 chars, which will fail if TASK_COMM_LEN is not 16. That looks like a constant in the kernel, so I think we're good.
viktormalik
pushed a commit
to BurntBrunch/bpftrace
that referenced
this issue
May 24, 2023
As noted in bpftrace#2402, usdt flakiness was made better by 508538a but not fully fixed. This commit is what I should have done all along: it allows the test runner to parse and wait for multiple BEFORE clauses and thus ensures the processes have started before the test runs. There are two minor changes: 1. The check for child processes is now `ps --ppid` based to eventually allow parallel process runs in the same environment. That requires to use `ps` from the `procps` package on Alpine as the default BusyBox one doesn't have the `--ppid` option. 2. Because of the `ps` usage, the name check is now truncated to 15 chars, which will fail if TASK_COMM_LEN is not 16. That looks like a constant in the kernel, so I think we're good.
danobi
pushed a commit
that referenced
this issue
May 24, 2023
As noted in #2402, usdt flakiness was made better by 508538a but not fully fixed. This commit is what I should have done all along: it allows the test runner to parse and wait for multiple BEFORE clauses and thus ensures the processes have started before the test runs. There are two minor changes: 1. The check for child processes is now `ps --ppid` based to eventually allow parallel process runs in the same environment. That requires to use `ps` from the `procps` package on Alpine as the default BusyBox one doesn't have the `--ppid` option. 2. Because of the `ps` usage, the name check is now truncated to 15 chars, which will fail if TASK_COMM_LEN is not 16. That looks like a constant in the kernel, so I think we're good.
jordalgo
added
the
tests
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
What reproduces the bug?
Trigger CI on the master branch.
https://github.com/iovisor/bpftrace/actions/runs/3343605763/jobs/5536998903
https://github.com/iovisor/bpftrace/actions/runs/3341593002/jobs/5532960039
The text was updated successfully, but these errors were encountered: