From: yite.gu <yitegu0@gmail.com>
Date: Thu, 18 Apr 2024 07:51:11 +0000 (+0800)
Subject: osd: add watch ping timeout count in osd
X-Git-Tag: v20.0.0~1372^2
X-Git-Url: http://git-server-git.apps.pok.os.sepia.ceph.com/?a=commitdiff_plain;h=34b086e702f45a500b0d112c129fa890ef6c9f59;p=ceph.git

osd: add watch ping timeout count in osd

For example, rbd send a watch ping to the header object
every 5 seconds to keep watch, if the primary OSD is unable
to receive the watch ping of the header object due to rbd
network interruption, this means that rbd's I/O has already been
hang. This way, we can quickly detect disconnection rbds on the osd,
and reflected in metrics.

Signed-off-by: Yite Gu <yitegu0@gmail.com>
---

diff --git a/src/osd/PrimaryLogPG.cc b/src/osd/PrimaryLogPG.cc
index 62b8aad27d5b..df25bbc96016 100644
--- a/src/osd/PrimaryLogPG.cc
+++ b/src/osd/PrimaryLogPG.cc
@@ -11826,6 +11826,10 @@ void PrimaryLogPG::handle_watch_timeout(WatchRef watch)
   oi.watchers.erase(make_pair(watch->get_cookie(),
 			      watch->get_entity()));
 
+  osd->logger->inc(l_osd_watch_timeouts);
+  dout(3) << __func__ << " watcher " << watch->get_peer_addr()
+	  << " object " << obc->obs.oi.soid << dendl;
+
   list<watch_disconnect_t> watch_disconnects = {
     watch_disconnect_t(watch->get_cookie(), watch->get_entity(), true)
   };
diff --git a/src/osd/osd_perf_counters.cc b/src/osd/osd_perf_counters.cc
index 30f0ba531665..a37fcbc29ba7 100644
--- a/src/osd/osd_perf_counters.cc
+++ b/src/osd/osd_perf_counters.cc
@@ -337,6 +337,10 @@ PerfCounters *build_osd_logger(CephContext *cct) {
   osd_plb.add_u64_counter_histogram(
       l_osd_scrub_reservation_dur_hist, "scrub_resrv_repnum_vs_duration",
       rsrv_hist_x_axis_config, rsrv_hist_y_axis_config, "Histogram of scrub replicas reservation duration");
+  osd_plb.add_u64_counter(
+  l_osd_watch_timeouts, "watch_timeouts",
+  "Number of watches that timed out or were blocklisted",
+  NULL, PerfCountersBuilder::PRIO_USEFUL);
 
   return osd_plb.create_perf_counters();
 }
diff --git a/src/osd/osd_perf_counters.h b/src/osd/osd_perf_counters.h
index 00127dd7ff5d..9ecbbd2368dc 100644
--- a/src/osd/osd_perf_counters.h
+++ b/src/osd/osd_perf_counters.h
@@ -136,6 +136,8 @@ enum {
   // are labeled, and histograms do not fully support labels.
   l_osd_scrub_reservation_dur_hist,
 
+  l_osd_watch_timeouts,
+
   l_osd_last,
 };