-
Notifications
You must be signed in to change notification settings - Fork 374
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
improve timeout_idle documentation wrt H2 #2927
Comments
FTR the meaning of |
With #2980 done, we now take timeout_idle into account also for h/2. I'll close this ticket now. |
Sorry - I noticed #2980 did not update the doc. I'll reopen for now |
Are you sure? I don't remember the specifics but I worked on different timeouts: idle_send_timeout and send_timeout, not timeout_idle. I think we are back to @nigoroll's comment. |
I'm not sure how @nigoroll feels about it but I think we shouldn't discuss this question today. As requested in #2983 (comment) we worked on bringing consistency to timeouts, and while we did not specifically address h2 we have a VIP draft that we may submit to the wiki to be discussed next bugwash. A side effect of this draft is that it should help us define what |
I agree with @dridi |
@dridi and myself still need to finish the timeout-proposal we have already spent quite some time on, but which we did not finish yet. This ticket seems to serve no real purpose, other than reminding me once again that we need to. |
Timed out. |
Exceptionally, one existing VTC was migrated to the new name immediately since it relied on debug logs containing 'idle_send_timeout'. Turning the timeout to an interrupt, I tried to update the parameter description to better convey what it does. This does not change the current behavior, but I tried to highlight what is wrong with this timeout and the mismatch between its purpose for HTTP/1 and h2. Depending on where we look at the documentation, between the parameter or VCL variables, we get a slightly different description. None of them are accurate, but the consensus seems to have always been the parameter description from before h2 or the 'sess.idle_send_timeout' variable. For session-level timeouts, we may consider an h2_send_timeout parameter in the future to distinguish between socket writes and underlying h2 streams timing out on a condvar. This h2_send_timeout parameter would map to SO_SNDTIMEO for h2 client connections. Refs varnishcache#2927
Exceptionally, one existing VTC was migrated to the new name immediately since it relied on debug logs containing 'idle_send_timeout'. Turning the timeout to an interrupt, I tried to update the parameter description to better convey what it does. This does not change the current behavior, but I tried to highlight what is wrong with this timeout and the mismatch between its purpose for HTTP/1 and h2. Depending on where we look at the documentation, between the parameter or VCL variables, we get a slightly different description. None of them are accurate, but the consensus seems to have always been the parameter description from before h2 or the 'sess.idle_send_timeout' variable. For session-level timeouts, we may consider an h2_send_timeout parameter in the future to distinguish between socket writes and underlying h2 streams timing out on a condvar. This h2_send_timeout parameter would map to SO_SNDTIMEO for h2 client connections. Refs varnishcache#2927
Exceptionally, one existing VTC was migrated to the new name immediately since it relied on debug logs containing 'idle_send_timeout'. Turning the timeout to an interrupt, I tried to update the parameter description to better convey what it does. This does not change the current behavior, but I tried to highlight what is wrong with this timeout and the mismatch between its purpose for HTTP/1 and h2. Depending on where we look at the documentation, between the parameter or VCL variables, we get a slightly different description. None of them are accurate, but the consensus seems to have always been the parameter description from before h2 or the 'sess.idle_send_timeout' variable. For session-level timeouts, we may consider an h2_send_timeout parameter in the future to distinguish between socket writes and underlying h2 streams timing out on a condvar. This h2_send_timeout parameter would map to SO_SNDTIMEO for h2 client connections. Refs varnishcache#2927
In varnishadm 6.1, documentation of timeout_idle says:
It appears that it's "particularly relevant" for HTTP/2 too ! I would says even more, as there's no more connection header in HTTP/2 ...
The text was updated successfully, but these errors were encountered: