tests/osd: creating a Teuthology test re missing SnapMapper entries

The test (in the standalone/scrub suite) verifies that the scrubber
detects (and issues a cluster-log error) whenever a mapping entry
("SNA_") is missing in the SnapMapper DB.

Specifically, here the entry is corrupted - shortened as per
https://tracker.ceph.com/issues/56147.

Signed-off-by: Ronen Friedman <rfriedma@redhat.com>

diff --git a/qa/standalone/scrub/osd-mapper.sh b/qa/standalone/scrub/osd-mapper.sh
new file mode 100755 (executable)
index 0000000..989271b
--- /dev/null
+++ b/qa/standalone/scrub/osd-mapper.sh
@@ -0,0 +1,158 @@
+#!/usr/bin/env bash
+# -*- mode:text; tab-width:8; c-basic-offset:2; indent-tabs-mode:t -*-
+# vim: ts=8 sw=2 smarttab
+#
+# test the handling of a corrupted SnapMapper DB by Scrub
+
+source $CEPH_ROOT/qa/standalone/ceph-helpers.sh
+source $CEPH_ROOT/qa/standalone/scrub/scrub-helpers.sh
+
+function run() {
+  local dir=$1
+  shift
+
+  export CEPH_MON="127.0.0.1:7144" # git grep '\<7144\>' : there must be only one
+  export CEPH_ARGS
+  CEPH_ARGS+="--fsid=$(uuidgen) --auth-supported=none "
+  CEPH_ARGS+="--mon-host=$CEPH_MON "
+
+  export -n CEPH_CLI_TEST_DUP_COMMAND
+  local funcs=${@:-$(set | sed -n -e 's/^\(TEST_[0-9a-z_]*\) .*/\1/p')}
+  for func in $funcs ; do
+    setup $dir || return 1
+    $func $dir || return 1
+    teardown $dir || return 1
+  done
+}
+
+# one clone & multiple snaps (according to the number of parameters)
+function make_a_clone()
+{
+  #turn off '-x' (but remember previous state)
+  local saved_echo_flag=${-//[^x]/}
+  set +x
+  local pool=$1
+  local obj=$2
+  echo $RANDOM | rados -p $pool put $obj - || return 1
+  shift 2
+  for snap in $@ ; do
+    rados -p $pool mksnap $snap || return 1
+  done
+  if [[ -n "$saved_echo_flag" ]]; then set -x; fi
+}
+
+function TEST_truncated_sna_record() {
+    local dir=$1
+    local -A cluster_conf=(
+        ['osds_num']="3" 
+        ['pgs_in_pool']="4"
+        ['pool_name']="test"
+    )
+
+    local extr_dbg=1
+    (( extr_dbg > 1 )) && echo "Dir: $dir"
+    standard_scrub_cluster $dir cluster_conf
+    ceph tell osd.* config set osd_stats_update_period_not_scrubbing "1"
+    ceph tell osd.* config set osd_stats_update_period_scrubbing "1"
+
+    local osdn=${cluster_conf['osds_num']}
+    local poolid=${cluster_conf['pool_id']}
+    local poolname=${cluster_conf['pool_name']}
+    local objname="objxxx"
+
+    # create an object and clone it
+    make_a_clone $poolname $objname snap01 snap02 || return 1
+    make_a_clone $poolname $objname snap13 || return 1
+    make_a_clone $poolname $objname snap24 snap25 || return 1
+    echo $RANDOM | rados -p $poolname put $objname - || return 1
+
+    #identify the PG and the primary OSD
+    local pgid=`ceph --format=json-pretty osd map $poolname $objname | jq -r '.pgid'`
+    local osd=`ceph --format=json-pretty osd map $poolname $objname | jq -r '.up[0]'`
+    echo "pgid is $pgid (primary: osd.$osd)"
+    # turn on the publishing of test data in the 'scrubber' section of 'pg query' output
+    set_query_debug $pgid
+
+    # verify the existence of these clones
+    (( extr_dbg >= 1 )) && rados --format json-pretty -p $poolname listsnaps $objname
+
+    # scrub the PG
+    ceph pg $pgid deep_scrub || return 1
+
+    # we aren't just waiting for the scrub to terminate, but also for the
+    # logs to be published
+    sleep 3
+    ceph pg dump pgs
+    until grep -a -q -- "event: --^^^^---- ScrubFinished" $dir/osd.$osd.log ; do
+        sleep 0.2
+    done
+
+    ceph pg dump pgs
+    ceph osd set noscrub || return 1
+    ceph osd set nodeep-scrub || return 1
+    sleep 5
+    grep -a -q -v "ERR" $dir/osd.$osd.log || return 1
+
+    # kill the OSDs
+    kill_daemons $dir TERM osd || return 1
+
+    (( extr_dbg >= 2 )) && ceph-kvstore-tool bluestore-kv $dir/0 dump "p"
+    (( extr_dbg >= 2 )) && ceph-kvstore-tool bluestore-kv $dir/2 dump "p" | grep -a SNA_
+    (( extr_dbg >= 2 )) && grep -a SNA_ /tmp/oo2.dump
+    (( extr_dbg >= 2 )) && ceph-kvstore-tool bluestore-kv $dir/2 dump p 2> /dev/null
+
+    for sdn in $(seq 0 $(expr $osdn - 1))
+    do
+        kvdir=$dir/$sdn
+        echo "corrupting the SnapMapper DB of osd.$sdn (db: $kvdir)"
+        (( extr_dbg >= 3 )) && ceph-kvstore-tool bluestore-kv $kvdir dump "p"
+
+        # truncate the 'mapping' (SNA_) entry corresponding to the snap13 clone
+        KY=`ceph-kvstore-tool bluestore-kv $kvdir dump p 2> /dev/null | grep -a -e 'SNA_[0-9]_0000000000000003_000000000000000' \
+            | awk -e '{print $2;}'`
+        (( extr_dbg >= 1 )) && echo "SNA key: $KY" | cat -v
+
+        tmp_fn1=`mktemp -p /tmp --suffix="_the_val"`
+        (( extr_dbg >= 1 )) && echo "Value dumped in: $tmp_fn1"
+        ceph-kvstore-tool bluestore-kv $kvdir get p "$KY" out $tmp_fn1 2> /dev/null
+        (( extr_dbg >= 2 )) && od -xc $tmp_fn1
+
+        NKY=${KY:0:-30}
+        ceph-kvstore-tool bluestore-kv $kvdir rm "p" "$KY" 2> /dev/null
+        ceph-kvstore-tool bluestore-kv $kvdir set "p" "$NKY" in $tmp_fn1 2> /dev/null
+
+        (( extr_dbg >= 1 )) || rm $tmp_fn1
+    done
+
+    orig_osd_args=" ${cluster_conf['osd_args']}"
+    orig_osd_args=" $(echo $orig_osd_args)"
+    (( extr_dbg >= 2 )) && echo "Copied OSD args: /$orig_osd_args/ /${orig_osd_args:1}/"
+    for sdn in $(seq 0 $(expr $osdn - 1))
+    do
+      CEPH_ARGS="$CEPH_ARGS $orig_osd_args" activate_osd $dir $sdn
+    done
+    sleep 1
+
+    for sdn in $(seq 0 $(expr $osdn - 1))
+    do
+      timeout 60 ceph tell osd.$sdn version
+    done
+
+    # when scrubbing now - we expect the scrub to emit a cluster log ERR message regarding SnapMapper internal inconsistency
+    ceph osd unset nodeep-scrub || return 1
+    ceph osd unset noscrub || return 1
+
+    # what is the primary now?
+    local cur_prim=`ceph --format=json-pretty osd map $poolname $objname | jq -r '.up[0]'`
+    ceph pg dump pgs
+    sleep 2
+    ceph pg $pgid deep_scrub || return 1
+    sleep 5
+    ceph pg dump pgs
+    (( extr_dbg >= 1 )) && grep -a "ERR" $dir/osd.$cur_prim.log
+    grep -a -q "ERR" $dir/osd.$cur_prim.log || return 1
+}
+
+
+
+main osd-mapper "$@"

diff --git a/qa/standalone/scrub/scrub-helpers.sh b/qa/standalone/scrub/scrub-helpers.sh
index c7f48150cc1d435e361c40a62c5d54ff84221bd9..38677cf756657029c87061a0d56fa6b027cdca32 100644 (file)
--- a/qa/standalone/scrub/scrub-helpers.sh
+++ b/qa/standalone/scrub/scrub-helpers.sh
@@ -243,7 +243,7 @@ function standard_scrub_cluster() {
 
     for osd in $(seq 0 $(expr $OSDS - 1))
     do
-      run_osd $dir $osd $ceph_osd_args || return 1
+      run_osd $dir $osd $(echo $ceph_osd_args) || return 1
     done
 
     create_pool $poolname $pg_num $pg_num
@@ -254,6 +254,7 @@ function standard_scrub_cluster() {
     name_n_id=`ceph osd dump | awk '/^pool.*'$poolname'/ { gsub(/'"'"'/," ",$3); print $3," ", $2}'`
     echo "standard_scrub_cluster: $debug_msg: test pool is $name_n_id"
     args['pool_id']="${name_n_id##* }"
+    args['osd_args']=$ceph_osd_args
     if [[ -n "$saved_echo_flag" ]]; then set -x; fi
 }
 

diff --git a/src/osd/scrubber/scrub_backend.cc b/src/osd/scrubber/scrub_backend.cc
index 1f7c734a6366edd0c53fd779bbe3a8fc0c61a9ea..e12eb106d14d7e883166bc670043e247f36e0b21 100644 (file)
--- a/src/osd/scrubber/scrub_backend.cc
+++ b/src/osd/scrubber/scrub_backend.cc
@@ -1892,13 +1892,15 @@ std::optional<snap_mapper_fix_t> ScrubBackend::scan_object_snaps(
 }
 
 /*
- * Process:
  * Building a map of objects suitable for snapshot validation.
- * The data in m_cleaned_meta_map is the leftover partial items that need to
- * be completed before they can be processed.
  *
- * Snapshots in maps precede the head object, which is why we are scanning
- * backwards.
+ * We are moving all "full" clone sets, i.e. the head and (preceding it, as
+ * snapshots precede the head entry) the clone entries, into 'for_meta_scrub'.
+ * That collection, not containing partial items, will be scrubbed by
+ * scrub_snapshot_metadata().
+ *
+ * What's left in m_cleaned_meta_map is the leftover partial items that need to
+ * be completed before they can be processed.
  */
 ScrubMap ScrubBackend::clean_meta_map(ScrubMap& cleaned, bool max_reached)
 {