From: Radosław Zarzyński <rzarzyns@redhat.com>
Date: Wed, 15 Jun 2022 15:19:30 +0000 (+0200)
Subject: osd: trim dups regardless of their versions
X-Git-Tag: v18.0.0~161^2~6
X-Git-Url: http://git.apps.os.sepia.ceph.com/?a=commitdiff_plain;h=aada08acde7a05ad769bb7a886ebcece628d522c;p=ceph-ci.git

osd: trim dups regardless of their versions

Fixes: https://tracker.ceph.com/issues/53729
Signed-off-by: Radoslaw Zarzynski <rzarzyns@redhat.com>
---

diff --git a/src/osd/PGLog.cc b/src/osd/PGLog.cc
index d7ab12c01df..c61e819f908 100644
--- a/src/osd/PGLog.cc
+++ b/src/osd/PGLog.cc
@@ -131,10 +131,18 @@ void PGLog::IndexedLog::trim(
     }
   }
 
-  while (!dups.empty()) {
+  // we can hit an inflated `dups` b/c of https://tracker.ceph.com/issues/53729
+  // the idea is to slowly trim them over a prolonged period of time and mix
+  // omap deletes with writes (if we're here, a new log entry got added) to
+  // neither: 1) blow size of single Transaction nor 2) generate-n-accumulate
+  // large amount of tombstones in BlueStore's RocksDB.
+  // if trimming immediately is a must, then the ceph-objectstore-tool is
+  // the way to go.
+  const size_t max_dups = cct->_conf->osd_pg_log_dups_tracked;
+  for (size_t max_dups_to_trim = cct->_conf->osd_pg_log_trim_max;
+       max_dups_to_trim > 0 && dups.size() > max_dups;
+       max_dups_to_trim--) {
     const auto& e = *dups.begin();
-    if (e.version.version >= earliest_dup_version)
-      break;
     lgeneric_subdout(cct, osd, 20) << "trim dup " << e << dendl;
     if (trimmed_dups)
       trimmed_dups->insert(e.get_key_name());