Poul-Henning Kamp
Poul-Henning Kamp
Sorry, but I see no reason for adding this complexity.
I understand the issue and the motivation for this proposal, but I am still not convinced it is something we should expend source code lines on.
I am really not happy about the cool-list, primarily because it is optimistic. I tried adding director refcounts cheaply some weeks ago, but it became messy so I gave up....
Is this ticket serving any purpose any more ?
bugwash consensus: Go for backing out optimization. Document with 200/304 data for posterity.
Discussed at bugwash. Uncertain if `EINTR` means "kernel clock stepped" so you may want to reevaluate this socket in re: timeouts or if some signal was actually sent. Handling options...
100% with @gquintard here, `beresp.is_error` is nogo. One could argue that being on the waiting list should be subject to the `first_byte_timeout`. Problem is we dont know the number, precisely...
Not against it. Any idea how many places in the code it could have been used already ?
Split in two (possibly 3) subjects: 1. Does being more informative, ie: 502 vs. 503) responses require VCL bump to 4.2 ? A: YES. 2. Exactly what should be returned...
Ok, here is my decision, relevant for #3017 as well: We will not change the current 503 behavior in VCL4.1 We will introduce this new behavior in VCL4.2 We can...