MINOR: stats: report the number of times the global maxconn was reached

As discussed a few times over the years, it's quite difficult to know
how often we stop accepting connections because the global maxconn was
reached. This is not easy to know because when we reach the limit we
stop accepting but we don't know if incoming connections are pending,
so it's not possible to know how many were delayed just because of this.
However, an interesting equivalent metric consist in counting the number
of times an accepted incoming connection resulted in the limit being
reached. I.e. "we've accepted the last one for now". That doesn't imply
any other one got delayed but it's a factual indicator that something
might have been delayed. And by counting the number of such events, it
becomes easier to know whether some limits need to be adjusted because
they're reached often, or if it's exceptionally rare.

The metric is reported as a counter in show info and on the stats page
in the info section right next to "maxconn".
This commit is contained in:
Willy Tarreau 2023-05-11 13:51:31 +02:00
parent 3c4a297d2b
commit 9615102b01
4 changed files with 16 additions and 2 deletions

View File

@ -220,6 +220,9 @@ extern struct accept_queue_ring accept_queue_rings[MAX_THREADS] __attribute__((a
extern const char* li_status_st[LI_STATE_COUNT];
enum li_status get_li_status(struct listener *l);
/* number of times an accepted connection resulted in maxconn being reached */
extern ullong maxconn_reached;
static inline uint accept_queue_ring_len(const struct accept_queue_ring *ring)
{
uint idx, head, tail, len;

View File

@ -344,6 +344,7 @@ enum info_field {
INF_START_TIME_SEC,
INF_TAINTED,
INF_WARNINGS,
INF_MAXCONN_REACHED,
/* must always be the last one */
INF_TOTAL_FIELDS

View File

@ -47,6 +47,8 @@ struct bind_kw_list bind_keywords = {
/* list of the temporarily limited listeners because of lack of resource */
static struct mt_list global_listener_queue = MT_LIST_HEAD_INIT(global_listener_queue);
static struct task *global_listener_queue_task;
/* number of times an accepted connection resulted in maxconn being reached */
ullong maxconn_reached = 0;
__decl_thread(static HA_RWLOCK_T global_listener_rwlock);
/* listener status for stats */
@ -1164,6 +1166,12 @@ void listener_accept(struct listener *l)
_HA_ATOMIC_INC(&activity[tid].accepted);
/* count the number of times an accepted connection resulted in
* maxconn being reached.
*/
if (unlikely(_HA_ATOMIC_LOAD(&actconn) + 1 >= global.maxconn))
_HA_ATOMIC_INC(&maxconn_reached);
/* past this point, l->bind_conf->accept() will automatically decrement
* l->nbconn, feconn and actconn once done. Setting next_*conn=0
* allows the error path not to rollback on nbconn. It's more

View File

@ -157,6 +157,7 @@ const struct name_desc info_fields[INF_TOTAL_FIELDS] = {
[INF_BUILD_INFO] = { .name = "Build info", .desc = "Build info" },
[INF_TAINTED] = { .name = "Tainted", .desc = "Experimental features used" },
[INF_WARNINGS] = { .name = "TotalWarnings", .desc = "Total warnings issued" },
[INF_MAXCONN_REACHED] = { .name = "MaxconnReached", .desc = "Number of times an accepted connection resulted in Maxconn being reached" },
};
const struct name_desc stat_fields[ST_F_TOTAL_FIELDS] = {
@ -3603,7 +3604,7 @@ static void stats_dump_html_info(struct stconn *sc, struct uri_auth *uri)
"<p><b>pid = </b> %d (process #%d, nbproc = %d, nbthread = %d)<br>\n"
"<b>uptime = </b> %dd %dh%02dm%02ds; warnings = %u<br>\n"
"<b>system limits:</b> memmax = %s%s; ulimit-n = %d<br>\n"
"<b>maxsock = </b> %d; <b>maxconn = </b> %d; <b>maxpipes = </b> %d<br>\n"
"<b>maxsock = </b> %d; <b>maxconn = </b> %d; <b>reached = </b> %llu; <b>maxpipes = </b> %d<br>\n"
"current conns = %d; current pipes = %d/%d; conn rate = %d/sec; bit rate = %.3f %cbps<br>\n"
"Running tasks: %d/%d; idle = %d %%<br>\n"
"</td><td align=\"center\" nowrap>\n"
@ -3641,7 +3642,7 @@ static void stats_dump_html_info(struct stconn *sc, struct uri_auth *uri)
global.rlimit_memmax ? ultoa(global.rlimit_memmax) : "unlimited",
global.rlimit_memmax ? " MB" : "",
global.rlimit_nofile,
global.maxsock, global.maxconn, global.maxpipes,
global.maxsock, global.maxconn, HA_ATOMIC_LOAD(&maxconn_reached), global.maxpipes,
actconn, pipes_used, pipes_used+pipes_free, read_freq_ctr(&global.conn_per_sec),
bps >= 1000000000UL ? (bps / 1000000000.0) : bps >= 1000000UL ? (bps / 1000000.0) : (bps / 1000.0),
bps >= 1000000000UL ? 'G' : bps >= 1000000UL ? 'M' : 'k',
@ -4740,6 +4741,7 @@ int stats_fill_info(struct field *info, int len, uint flags)
info[INF_TAINTED] = mkf_str(FO_STATUS, chunk_newstr(out));
info[INF_WARNINGS] = mkf_u32(FN_COUNTER, HA_ATOMIC_LOAD(&tot_warnings));
info[INF_MAXCONN_REACHED] = mkf_u32(FN_COUNTER, HA_ATOMIC_LOAD(&maxconn_reached));
chunk_appendf(out, "%#x", get_tainted());
return 1;