OSD: add recovery_wq suicide timeout

Signed-off-by: Samuel Just <sjust@redhat.com>

diff --git a/src/common/config_opts.h b/src/common/config_opts.h
index 1183df72e3e56ecb018513503bf0011a3c81e565..0bb6db68524b2e4b4238a9f42452ef6e4e019ee4 100644 (file)
--- a/src/common/config_opts.h
+++ b/src/common/config_opts.h
@@ -584,6 +584,7 @@ OPTION(osd_backfill_scan_min, OPT_INT, 64)
 OPTION(osd_backfill_scan_max, OPT_INT, 512)
 OPTION(osd_op_thread_timeout, OPT_INT, 15)
 OPTION(osd_recovery_thread_timeout, OPT_INT, 30)
+OPTION(osd_recovery_thread_suicide_timeout, OPT_INT, 300)
 OPTION(osd_recovery_sleep, OPT_FLOAT, 0)         // seconds to sleep between recovery ops
 OPTION(osd_snap_trim_thread_timeout, OPT_INT, 60*60*1)
 OPTION(osd_snap_trim_sleep, OPT_FLOAT, 0)

diff --git a/src/osd/OSD.cc b/src/osd/OSD.cc
index b6e04023a9935c788f9e1578fe137bdfd24775e5..4b8281c80f06e425b9a85b8a4747d0de64c89c2d 100644 (file)
--- a/src/osd/OSD.cc
+++ b/src/osd/OSD.cc
@@ -1521,7 +1521,11 @@ OSD::OSD(CephContext *cct_, ObjectStore *store_,
     cct->_conf->osd_command_thread_suicide_timeout,
     &command_tp),
   recovery_ops_active(0),
-  recovery_wq(this, cct->_conf->osd_recovery_thread_timeout, &recovery_tp),
+  recovery_wq(
+    this,
+    cct->_conf->osd_recovery_thread_timeout,
+    cct->_conf->osd_recovery_thread_suicide_timeout,
+    &recovery_tp),
   replay_queue_lock("OSD::replay_queue_lock"),
   snap_trim_wq(this, cct->_conf->osd_snap_trim_thread_timeout, &disk_tp),
   scrub_wq(this, cct->_conf->osd_scrub_thread_timeout, &disk_tp),

diff --git a/src/osd/OSD.h b/src/osd/OSD.h
index b90a0c9cb31863e2b1e63296646245f8c58c215e..b492ce58b5a851010ef6211617667c2d2e0b32e8 100644 (file)
--- a/src/osd/OSD.h
+++ b/src/osd/OSD.h
@@ -2023,8 +2023,8 @@ protected:
 
   struct RecoveryWQ : public ThreadPool::WorkQueue<PG> {
     OSD *osd;
-    RecoveryWQ(OSD *o, time_t ti, ThreadPool *tp)
-      : ThreadPool::WorkQueue<PG>("OSD::RecoveryWQ", ti, ti*10, tp), osd(o) {}
+    RecoveryWQ(OSD *o, time_t ti, time_t si, ThreadPool *tp)
+      : ThreadPool::WorkQueue<PG>("OSD::RecoveryWQ", ti, si, tp), osd(o) {}
 
     bool _empty() {
       return osd->recovery_queue.empty();