www/apache
1
0
Fork 0
mirror of https://github.com/apache/httpd.git synced 2026-10-23 22:49:01 +03:00
Code Activity

mod_proxy_beacon: recover pre-existing balancer members on graceful restart

Browse Source 
The per-URL member table (ctx->seen) lives in child-process memory while
balancer workers persist in shared memory, so after a graceful restart
ctx->seen is empty but the balancer still holds the announced members.
balancer_manage() refuses to re-add an existing worker, so the add failed,
m->added was never set, and a previously-evicted (DISABLED) member stayed
out of rotation forever.

On a failed add, look the worker up directly (balancer_manage itself
silently succeeds when it cannot find a worker, so the bare failure return
cannot distinguish "already present" from "full").  Treat an existing
worker as already added and fall through to the enable step; only a
genuinely full balancer keeps the throttled retry behavior.


git-svn-id: https://svn.apache.org/repos/asf/httpd/httpd/trunk@1938135 13f79535-47bb-0310-9956-ffa450edef68
This commit is contained in:
Jim Jagielski
2026-09-11 22:48:21 +00:00
parent 176aed0c65
commit 752003aa4d
1 changed files with 34 additions and 5 deletions
Download Patch File Download Diff File Expand all files Collapse all files

39
modules/proxy/mod_proxy_beacon.c
View File

@@ -688,6 +688,24 @@ static void beacon_log_throttled(beacon_ctx_t *ctx, apr_time_t now, const char *
}
}
/*
* Does url already exist as a member worker of the configured balancer?
* balancer_manage()/balancer_process_balancer_worker() silently succeed on an
* add of a worker it cannot find, so on a failed add we look the worker up
* directly to distinguish "already present" (recoverable) from "full" (not).
*/
static int beacon_worker_exists(beacon_ctx_t *ctx, apr_pool_t *pool,
const char *url)
{
proxy_server_conf *conf =
ap_get_module_config(ctx->s->module_config, &proxy_module);
proxy_balancer *bsel = ap_proxy_get_balancer(pool, conf,
apr_pstrcat(pool, BALANCER_PREFIX, ctx->balancer_name, NULL), 0);
return bsel
&& ap_proxy_get_worker(pool, bsel, conf, url) != NULL;
}
/*
* Attempt to add (and enable) url as a member of the configured balancer,
* reusing balancer_manage(). Two calls: add (the worker is created DISABLED by
@@ -723,11 +741,22 @@ static apr_status_t beacon_try_add(beacon_ctx_t *ctx, apr_pool_t *pool,
rv = ctx->manage_fn(r, params);
apr_pool_destroy(subp);
if (rv != APR_SUCCESS) {
/* Most likely the balancer has no free slots (grow it via ProxySet
* growth / BalancerGrowth). Throttled log; retried only after the
* backoff window, not on every announcement. */
beacon_log_throttled(ctx, now, "balancer add failed (full?)");
return rv;
/* A "worker already exists" report is the normal case after a graceful
* restart, when the balancer's members (persisted in shared memory)
* outlive this process's empty ctx->seen. balancer_manage() refuses to
* re-add an existing worker, so treat it as already-present and fall
* through to the enable step -- otherwise a previously-evicted member
* would stay DISABLED forever. A genuinely full balancer (no such
* worker, no free slot) is throttled and retried only after the backoff
* window, not on every announcement. */
if (!beacon_worker_exists(ctx, pool, url)) {
beacon_log_throttled(ctx, now, "balancer add failed (full?)");
return rv;
}
ap_log_error(APLOG_MARK, APLOG_NOTICE, 0, s,
APLOGNO(10589) "mod_proxy_beacon: backend %s already in "
"balancer://%s (recovered after restart)",
url, ctx->balancer_name);
}
/* Step 2: enable it (clear the DISABLED flag) so it serves traffic.