Mercurial > hg > nginx
changeset 4207:4fc91bae6f83
Better recheck of dead upstream servers.
Previously nginx used to mark backend again as live as soon as fail_timeout
passes (10s by default) since last failure. On the other hand, detecting
dead backend takes up to 60s (proxy_connect_timeout) in typical situation
"backend is down and doesn't respond to any packets". This resulted in
suboptimal behaviour in the above situation (up to 23% of requests were
directed to dead backend with default settings).
More detailed description of the problem may be found here (in Russian):
http://mailman.nginx.org/pipermail/nginx-ru/2011-August/042172.html
Fix is to only allow one request after fail_timeout passes, and
mark backend as "live" only if this request succeeds.
Note that with new code backend will not be marked "live" unless "check"
request is completed, and this may take a while in some specific workloads
(e.g. streaming). This is believed to be acceptable.
author | Maxim Dounin <mdounin@mdounin.ru> |
---|---|
date | Wed, 12 Oct 2011 14:22:48 +0000 |
parents | 1a94a56a4e5d |
children | 470462cfa31b |
files | src/http/modules/ngx_http_upstream_ip_hash_module.c src/http/ngx_http_upstream_round_robin.c src/http/ngx_http_upstream_round_robin.h |
diffstat | 3 files changed, 18 insertions(+), 8 deletions(-) [+] |
line wrap: on
line diff
--- a/src/http/modules/ngx_http_upstream_ip_hash_module.c Wed Oct 12 13:28:03 2011 +0000 +++ b/src/http/modules/ngx_http_upstream_ip_hash_module.c Wed Oct 12 14:22:48 2011 +0000 @@ -185,8 +185,8 @@ break; } - if (now - peer->accessed > peer->fail_timeout) { - peer->fails = 0; + if (now - peer->checked > peer->fail_timeout) { + peer->checked = now; break; } }
--- a/src/http/ngx_http_upstream_round_robin.c Wed Oct 12 13:28:03 2011 +0000 +++ b/src/http/ngx_http_upstream_round_robin.c Wed Oct 12 14:22:48 2011 +0000 @@ -443,8 +443,8 @@ break; } - if (now - peer->accessed > peer->fail_timeout) { - peer->fails = 0; + if (now - peer->checked > peer->fail_timeout) { + peer->checked = now; break; } @@ -491,8 +491,8 @@ break; } - if (now - peer->accessed > peer->fail_timeout) { - peer->fails = 0; + if (now - peer->checked > peer->fail_timeout) { + peer->checked = now; break; } @@ -663,15 +663,16 @@ return; } + peer = &rrp->peers->peer[rrp->current]; + if (state & NGX_PEER_FAILED) { now = ngx_time(); - peer = &rrp->peers->peer[rrp->current]; - /* ngx_lock_mutex(rrp->peers->mutex); */ peer->fails++; peer->accessed = now; + peer->checked = now; if (peer->max_fails) { peer->current_weight -= peer->weight / peer->max_fails; @@ -686,6 +687,14 @@ } /* ngx_unlock_mutex(rrp->peers->mutex); */ + + } else { + + /* mark peer live if check passed */ + + if (peer->accessed < peer->checked) { + peer->fails = 0; + } } rrp->current++;