[BUG] Segmentation fault at 0x0000000000000000 in rb_fiber_transfer
#113
Comments
|
Thanks for your report, I will take a look tomorrow. If I prepare a debug release, would you be able to test it? |
|
Of |
|
I have a couple of questions to help me understand the nature of the problem.
|
Ok, but keep in mind I'm here with some sort of SRE hat. I don't know the code using
Yeah, it's running inside Pitchfork, in a fairly gigantic application.
Yeah, unfortunately I don't know that, and as mentioned I can't just rely on the backtrace to point me at the calling code like I usually do for this sort of issues. All I know is that it's in a generic So I think it's a pool of HTTP connections? If I grep for results = Async do |task|
tasks = batch_entries.map do |batch_entry|
task.async do
run_with_cache(batch_entry.context, batch_entry.http_request)
end
end
tasks.map(&:wait)
end.wait Async do |task_scheduler|
tasks = scope_key_to_missing_fetch_results.map do |scope_key, missing_fetch_results|
task_scheduler.async do
perform_fetch(scope_key: scope_key, fetch_results: missing_fetch_results)
rescue StandardError => e
missing_fetch_results.each do |fetch_result|
fetch_result.error_result = e.message
end
log_fetch_error(e, message: "[REDACTED]")
end
end
tasks.map(&:wait)
end.waitThat seems to be all what it's used for. As for the organization of threads, as you can see in the crash report there's 16 Ruby threads, but also a whole lot of native threads started by gems like
Well, I don't know how often this codepath is called, but probably quite a lot, dozens of times a day isn't a whole lot for us, probably |
|
Thanks, that's helpful. |
|
The mention of |
|
It's not specific to graphql, it's also failing on some non graphql endpoints. |
Opening here first because I'm too unfamiliar with this to know whether it's a Ruby bug or a bug in
io-event. All I can say is thatrb_fiber_transferexpectsrb_current_execution_context()to return something, but it's called from a thread that doesn't have a current execution context.So I don't know if it's IO-event trying to resume a fiber from a wrong thread or wrong assumption in
rb_fiber_transfer.I unfortunately don't have a repro script, but this is happening on a fairly frequent basis (dozens of times a day).
The text was updated successfully, but these errors were encountered: