https://bz.apache.org/bugzilla/show_bug.cgi?id=65642

abernal <[email protected]> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 OS|                            |All

--- Comment #1 from abernal <[email protected]> ---
Root-caused this. It's not a regression, it's an intentional behavior change:
commit 1f6fd365c3 (Yann Ylavic, 2015-01-07, "mod_proxy: Don't put non
balancer-member workers in error state by default for connection or 500/503
errors, and honor status=+I for any error. PR 48388") added this to
proxy_util.c's worker creation:

  if (!balancer) {
      wshared->status |= PROXY_WORKER_IGNORE_ERRORS;
  }

Every worker that is not a member of a <Proxy balancer://...> (i.e. any plain
ProxyPass) gets PROXY_WORKER_IGNORE_ERRORS set at creation. Both
ap_proxy_connect_backend()'s and mod_proxy.c's "mark worker in error" logic are
gated on this flag not being set, so retry= has no effect at all by default for
a plain ProxyPass worker -- by design, since 2.4.17/2.4.18ish. That exactly
matches your own timeline (worked through 2.4.16, broken after).

Confirmed the documented opt-out restores the old behavior. I reproduced with a
real backend address that genuinely times out on connect:

ProxyPass "/svc/" "http://10.255.255.1:80/"; retry=5 connectiontimeout=2
status=-I

First request takes the full 2s (connect attempt), and a second request within
the retry=5 window now returns instantly (503) instead of re-attempting the
connection -- exactly the behavior you're asking for. Without "status=-I" both
requests take the full 2s every time.

So the fix for your setup is adding status=-I to the ProxyPass line. The real
gap I found is in the documentation: the "retry=" parameter description in
mod_proxy.xml doesn't mention that it has no effect for a non-balancer worker
unless status=-I is also set. Happy to send a small doc patch for that if it'd
help; let me know.

-- 
You are receiving this mail because:
You are the assignee for the bug.
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]

Reply via email to