Fix missing historyUUID in peer recovery when rolling upgrade 5.x to 6.3 #31506

dnhatn · 2018-06-21T14:32:22Z

Today we make sure that a 5.x index commit should have all required
commit tags in RecoveryTarget#cleanFiles method. The reason we do this
in RecoveryTarget#cleanFiles method because this is only needed in a
file-based recovery and we assume that #cleanFiles should be called in a
file-based recovery. However, this assumption is not valid if the index
is sealed (.i.e synced-flush). This incorrect assumption would prevent
users from rolling upgrade from 5.x to 6.3 if their index were sealed.

Closes #31482

Today we make sure that a 5.x index commit should have all required commit tags in RecoveryTarget#cleanFiles method. The reason we do this in RecoveryTarget#cleanFiles method because this is only needed in a file-based recovery and we assume that #cleanFiles should be called in a file-based recovery. However, this assumption is not valid if the index is sealed (.i.e synced-flush). This incorrect assumption would prevent users from rolling upgrade from 5.x to 6.3 if their index were sealed.

elasticmachine · 2018-06-21T14:32:24Z

Pinging @elastic/es-distributed

ywelsch

Left some nits but looks good o.w.

ywelsch · 2018-06-21T14:57:08Z

qa/full-cluster-restart/src/test/java/org/elasticsearch/upgrades/FullClusterRestartIT.java

+            createIndex(index, settings.build());
+            indexRandomDocuments(count, randomBoolean(), randomBoolean(),
+                n -> JsonXContent.contentBuilder().startObject().field("key", "value").endObject());
+            assertBusy(() -> {


Can we replace this assertBusy by an ensureGreen(index)?

ywelsch · 2018-06-21T14:58:24Z

qa/full-cluster-restart/src/test/java/org/elasticsearch/upgrades/FullClusterRestartIT.java

+     */
+    public void testRecoverySealedIndex() throws Exception {
+        int count;
+        if (runningAgainstOldCluster) {


maybe fold this if-else into the next one.

ywelsch · 2018-06-21T14:58:40Z

qa/full-cluster-restart/src/test/java/org/elasticsearch/upgrades/FullClusterRestartIT.java

@@ -892,6 +932,7 @@ public void testHistoryUUIDIsAdded() throws Exception {
            mappingsAndSettings.endObject();
            client().performRequest("PUT", "/" + index, Collections.emptyMap(),
                new StringEntity(Strings.toString(mappingsAndSettings), ContentType.APPLICATION_JSON));
+


remove extra line

ywelsch · 2018-06-21T15:01:28Z

qa/full-cluster-restart/src/test/java/org/elasticsearch/upgrades/FullClusterRestartIT.java

+     * Tests that a synced-flushed index is correctly recovered.
+     * This might be an edge-case from 5.x to 6.x since a 5.x index commit does not have all required 6.x commit tags.
+     */
+    public void testRecoverySealedIndex() throws Exception {


testRecoverSyncedFlushIndex

ywelsch · 2018-06-21T15:01:53Z

qa/rolling-upgrade/src/test/java/org/elasticsearch/upgrades/RecoveryIT.java

@@ -285,4 +286,27 @@ public void testSearchGeoPoints() throws Exception {
        }
    }

+    public void testRecoverySealedIndex() throws Exception {


testRecoverSyncedFlushIndex

ywelsch · 2018-06-21T15:02:12Z

qa/rolling-upgrade/src/test/java/org/elasticsearch/upgrades/RecoveryIT.java

@@ -285,4 +286,27 @@ public void testSearchGeoPoints() throws Exception {
        }
    }

+    public void testRecoverySealedIndex() throws Exception {
+        final String index = "recovery_a_sealed_index";


synced_flush_index

ywelsch · 2018-06-21T15:02:41Z

qa/rolling-upgrade/src/test/java/org/elasticsearch/upgrades/RecoveryIT.java

+                .put(SETTING_ALLOCATION_MAX_RETRY.getKey(), "0"); // fail faster
+            createIndex(index, settings.build());
+            indexDocs(index, 0, randomInt(5));
+            assertBusy(() -> {


same comment as above, replace by ensureGreen()?

dnhatn · 2018-06-21T15:54:00Z

@ywelsch Thanks for looking.

Can we replace this assertBusy by an ensureGreen(index)?

Unfortunately, we have to spin here. We fire a global checkpoint sync after the last op is completed. A synced-flush will fail if the global checkpoint is happening. I added a comment to explain this.

bleskes

LGTM. I wonder if #31291 is relevant (it is different because it doesn't fail to recover). I think it's worth while double checking that.

bleskes · 2018-06-21T18:58:26Z

qa/full-cluster-restart/src/test/java/org/elasticsearch/upgrades/FullClusterRestartIT.java

+     * Tests that a synced-flushed index is correctly recovered.
+     * This might be an edge-case from 5.x to 6.x since a 5.x index commit does not have all required 6.x commit tags.
+     */
+    public void testRecoverSyncedFlushIndex() throws Exception {


can't we fold this into testRecovery and randomly do a synced flush instead of a normal flush?

dnhatn · 2018-06-21T20:52:01Z

I wonder if #31291 is relevant (it is different because it doesn't fail to recover). I think it's worth while double checking that.

Sure, will do.

dnhatn · 2018-06-22T03:28:36Z

Thanks @ywelsch and @bleskes for reviewing!

…6.3 (#31506) Today we make sure that a 5.x index commit should have all required commit tags in RecoveryTarget#cleanFiles method. The reason we do this in RecoveryTarget#cleanFiles method because this is only needed in a file-based recovery and we assume that #cleanFiles should be called in a file-based recovery. However, this assumption is not valid if the index is sealed (.i.e synced-flush). This incorrect assumption would prevent users from rolling upgrade from 5.x to 6.3 if their index were sealed. Closes #31482

Although the master branch does not affect by #31482, it's helpful to have BWC tests that verify the peer recovery with a synced-flush index. This commit adds the bwc tests from #31506 to the master branch. Relates #31482 Relates #31506

dnhatn · 2018-06-23T00:18:28Z

Although the master branch is not affected by this issue, I think it's nice to have these BWC tests in the master branch. I added these tests to the master branch in 5115102.

* 6.x: Avoid sending duplicate remote failed shard requests (#31313) Add get field mappings to High Level REST API Client Relates to #27205 [DOCS] Updates Watcher examples for code testing (#31152) [DOCS] Move monitoring to docs folder (#31477) [DOCS] Fixes SQL docs in nav [DOCS] Move sql to docs IndexShard should not return null stats - empty stats or AlreadyCloseException if it's closed is better Clarify that IP range data can be specified in CIDR notation. (#31374) Remove some cases in FieldTypeLookupTests that are no longer relevant. (#31381) In NumberFieldType equals and hashCode, make sure that NumberType is taken into account. (#31514) fix repository update with the same settings but different type Revert "AwaitsFix FullClusterRestartIT#testRecovery" Upgrade to Lucene 7.4.0. (#31529) Avoid deprecation warning when running the ML datafeed extractor. (#31463) Retry synced-flush in FullClusterRestartIT#testRecovery Allow multiple unicast host providers (#31509) [ML] Add ML filter update API (#31437) AwaitsFix FullClusterRestartIT#testRecovery Fix missing historyUUID in peer recovery when rolling upgrade 5.x to 6.3 (#31506) Remove QueryCachingPolicy#ALWAYS_CACHE (#31451) Rename createNewTranslog to fileBasedRecovery (#31508) [DOCS] Add code snippet testing in more ML APIs (#31339) [DOCS] Remove fixed file from build.gradle [DOCS] Creates field and document level security overview (#30937) Test: Skip assertion on windows [DOCS] Move migration APIs to docs (#31473) Add a known issue for upgrading from 5.x to 6.3.0 (#31501) Return transport addresses from UnicastHostsProvider (#31426) Add Delete Snapshot High Level REST API Reload secure settings for plugins (#31481) [DOCS] Fix JDBC Maven client group/artifact ID

Although the master branch does not affect by #31482, it's helpful to have BWC tests that verify the peer recovery with a synced-flush index. This commit adds the bwc tests from #31506 to the master branch. Relates #31482 Relates #31506

bleskes · 2018-06-25T09:33:24Z

I added these tests to the master branch in 5115102.

+1

dnhatn added >bug >upgrade :Distributed/Recovery Anything around constructing a new shard, either from a local or a remote source. v6.4.0 v6.3.1 labels Jun 21, 2018

dnhatn requested review from bleskes and ywelsch June 21, 2018 14:32

ywelsch approved these changes Jun 21, 2018

View reviewed changes

Yannick feedback

b93cc20

bleskes approved these changes Jun 21, 2018

View reviewed changes

fold to the recovery test

814bcfa

Correct number_of_shards

fd89c62

dnhatn merged commit 9100a70 into elastic:6.x Jun 22, 2018

dnhatn deleted the fix-missing-historyuuid branch June 22, 2018 03:29

dnhatn added the backport pending label Jun 22, 2018

dnhatn mentioned this pull request Jun 22, 2018

5.6.10 to 6.3.0 rolling upgrade broken with 'commit doesn't contain history uuid' when a synced flush is performed #31482

Closed

dnhatn removed the backport pending label Jun 22, 2018

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Fix missing historyUUID in peer recovery when rolling upgrade 5.x to 6.3 #31506

Fix missing historyUUID in peer recovery when rolling upgrade 5.x to 6.3 #31506

dnhatn commented Jun 21, 2018 •

edited

Loading

elasticmachine commented Jun 21, 2018

ywelsch left a comment

ywelsch Jun 21, 2018

ywelsch Jun 21, 2018

ywelsch Jun 21, 2018

ywelsch Jun 21, 2018

ywelsch Jun 21, 2018

ywelsch Jun 21, 2018

ywelsch Jun 21, 2018

dnhatn commented Jun 21, 2018

bleskes left a comment

bleskes Jun 21, 2018

dnhatn commented Jun 21, 2018

dnhatn commented Jun 22, 2018

dnhatn commented Jun 23, 2018 •

edited

Loading

bleskes commented Jun 25, 2018

Fix missing historyUUID in peer recovery when rolling upgrade 5.x to 6.3 #31506

Fix missing historyUUID in peer recovery when rolling upgrade 5.x to 6.3 #31506

Conversation

dnhatn commented Jun 21, 2018 • edited Loading

elasticmachine commented Jun 21, 2018

ywelsch left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

dnhatn commented Jun 21, 2018

bleskes left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

dnhatn commented Jun 21, 2018

dnhatn commented Jun 22, 2018

dnhatn commented Jun 23, 2018 • edited Loading

bleskes commented Jun 25, 2018

dnhatn commented Jun 21, 2018 •

edited

Loading

dnhatn commented Jun 23, 2018 •

edited

Loading