-
Dridi Boukelmoune authored
When a session is continuously used, for example in a multi-tier setup, it can take a very long time before an HTTP/1 session runs out of tasks and unwinds back to the point where the worker might update its stats. This phenomenon is amplified by thread_stats_rate, so it might take even longer to get statistics published. This is famously observed with load testing campaigns where a certain set of VSCs may not increase at all until the very end of a load run. This change makes it possible for HTTP/1 sessions to at least publish their statistics into the pool every thread_stats_rate requests. To avoid a potential performance cost, this behavior (technically complying with the documentation) is guarded by a feature flag. The flag is generic so this behavior may be extended to other busy loops, present (if any) or future.
849c0844
Name |
Last commit
|
Last update |
---|---|---|
.. | ||
cache | ||
common | ||
fuzzers | ||
hash | ||
hpack | ||
http1 | ||
http2 | ||
mgt | ||
proxy | ||
storage | ||
waiter | ||
Makefile.am | ||
VSC_lck.vsc | ||
VSC_main.vsc | ||
VSC_mempool.vsc | ||
VSC_mgt.vsc | ||
VSC_sma.vsc | ||
VSC_smf.vsc | ||
VSC_smu.vsc | ||
VSC_vbe.vsc | ||
builtin.vcl | ||
flint.lnt | ||
flint.sh | ||
vclflint.lnt | ||
vclflint.sh |