osd/PG: fix last_complete re-calculation on splitting

We add hard-limit for pg_logs now, which means we might keep trimming
old log entries irrespective of pg's current missing_set. This as a
result can cause the last_complete pointer moving far ahead of the real
on-disk version (the oldest need of missing_set, for instance) the
corresponding pg should have on splitting:

```
2019-04-19 06:41:52.559247 7efd4725c700 10 osd.2 271 Splitting pg[5.6( v 270'943 lc 0'0 (238'300,270'943] local-lis/les=250/251 n=943 ec=223/223 lis/c 250/223 les/251/224/0 250/271/229) [5,2] r=1 lpr=271 pi=[223,271)/4 crt=270'943 unknown NOTIFY m=518 mbc={}] into 5.16
2019-04-19 06:41:52.561413 7efd4725c700 10 osd.2 pg_epoch: 271 pg[5.6( v 270'943 lc 238'300 (238'300,270'943] local-lis/les=250/251 n=943 ec=223/223 lis/c 250/223 c/f 251/224/0 250/271/229) [5,2] r=1 lpr=271 pi=[223,271)/4 crt=270'943 lcod 0'0 unknown NOTIFY m=261 mbc={}] release_backoffs [MIN,MAX)
```

For the above example, parent's last_complete cursor changed from **0'0** to
**238'300** directly due to the effort of trying to catch up the oldest
log entry changing when splitting was done. However, back into v12.2.9 primary
would still reference shard's last_complete field when trying to figure out all
possible locations of a currently missing object (see PG::MissingLoc::add_source_info):

```c++
  if (oinfo.last_complete < need) {
    if (omissing.is_missing(soid)) {
      ldout(pg->cct, 10) << "search_for_missing " << soid << " " << need
                         << " also missing on osd." << fromosd << dendl;
      continue;
    }
  }
```

Hence a wrongly calculated last_complete could then make primary mis-consider
that a specific shard might have the authoritative object it currently
looking for:

```
2019-04-19 06:41:52.904163 7fd4cfb5a700 10 osd.5 pg_epoch: 271 pg[5.6( v 270'943 lc 238'300 (238'300,270'943] local-lis/les=250/251 n=471 ec=223/223 lis/c 250/223 les/
c/f 251/224/0 250/271/229) [5,2] r=0 lpr=271 pi=[223,271)/4 crt=270'943 lcod 226'77 mlcod 0'0 peering m=16 mbc={}] proc_replica_log for osd.2: 5.6( v 270'943 lc 238'30
0 (238'300,270'943] local-lis/les=250/251 n=471 ec=223/223 lis/c 250/223 les/c/f 251/224/0 250/271/229) log((249'563,270'943], crt=270'943) missing(261 may_include_del
etes = 1)
2019-04-19 06:41:52.904645 7fd4cfb5a700 20 osd.5 pg_epoch: 271 pg[5.6( v 270'943 lc 238'300 (238'300,270'943] local-lis/les=250/251 n=471 ec=223/223 lis/c 250/223 les/
c/f 251/224/0 250/271/229) [5,2] r=0 lpr=271 pi=[223,271)/4 crt=270'943 lcod 226'77 mlcod 0'0 peering m=16 mbc={}]  after missing 5:624c3a7a:::benchmark_data_smithi190
_39968_object1382:head need 226'110 have 0'0
2019-04-19 06:41:53.567820 7fd4d035b700 10 osd.5 pg_epoch: 272 pg[5.6( v 270'943 lc 0'0 (238'300,270'943] local-lis/les=271/272 n=471 ec=223/223 lis/c 250/223 les/c/f
251/224/0 250/271/229) [5,2] r=0 lpr=271 pi=[223,271)/4 crt=270'943 lcod 226'77 mlcod 0'0 unknown m=16 u=13 mbc={255={(1+0)=220,(2+0)=28}}] search_for_missing 5:624c3a
7a:::benchmark_data_smithi190_39968_object1382:head 226'110 is on osd.2
```

note that ```5:624c3a7a:::benchmark_data_smithi190_39968_object1382:head 226'110```
was actually missing on both primary and shard osd.2 whereas primary insisted that
object should exist on shard osd.2!

https://github.com/ceph/ceph/pull/26175 posted an indirect fix
for the above problem by ignoring last_complete when checking the missing set,
but it should generally make more sense to fill in the last_complete field correctly
whenever possible.
Hence coming this additional fix.

Fixes: http://tracker.ceph.com/issues/26958
Signed-off-by: xie xingguo <xie.xingguo@zte.com.cn>

diff --git a/src/osd/PG.cc b/src/osd/PG.cc
index ffdce25ab225ea4f26b57ab5c91b2a45f473a733..4f1872cf493636775db7cedeed56c5652af3fc89 100644 (file)
--- a/src/osd/PG.cc
+++ b/src/osd/PG.cc
@@ -2624,10 +2624,9 @@ void PG::split_into(pg_t child_pgid, PG *child, unsigned split_bits)
   info.log_tail = pg_log.get_tail();
   child->info.log_tail = child->pg_log.get_tail();
 
-  if (info.last_complete < pg_log.get_tail())
-    info.last_complete = pg_log.get_tail();
-  if (child->info.last_complete < child->pg_log.get_tail())
-    child->info.last_complete = child->pg_log.get_tail();
+  // reset last_complete, we might have modified pg_log & missing above
+  pg_log.reset_complete_to(&info);
+  child->pg_log.reset_complete_to(&child->info);
 
   // Info
   child->info.history = info.history;

diff --git a/src/osd/PGLog.h b/src/osd/PGLog.h
index 6b492785e9e1e4fc7efcf4e0a1e02db2b9dfebab..f02b4c4c23d199dcdd02ac1049ac290762d9d34c 100644 (file)
--- a/src/osd/PGLog.h
+++ b/src/osd/PGLog.h
@@ -797,6 +797,8 @@ public:
   }
 
   void reset_complete_to(pg_info_t *info) {
+    if (log.log.empty()) // caller is split_into()
+      return;
     log.complete_to = log.log.begin();
     ceph_assert(log.complete_to != log.log.end());
     auto oldest_need = missing.get_oldest_need();