CCS: don't proxy requests for already connected node by javanna · Pull Request #31273 · elastic/elasticsearch

javanna · 2018-06-12T13:41:13Z

Cross-cluster search selects a subset of nodes for each remote cluster
and sends requests only to them, which will act as a proxy and properly
redirect such requests to the target nodes that hold the relevant data.
What happens today is that every time we send a request to a remote
cluster, it will be sent to the next node in the proxy list
(in round-robin fashion), regardless of whether the target node is
already amongst the ones that we are connected to. In case for instance
we need to send a shard search request to a data node that's also one of
the selected proxy nodes, we may end up sending the request to it
through one of the other proxy nodes.
This commit optimizes this case to make sure that whenever we are
already connected to a remote node, we will send a direct request rather
than using the next proxy node.
There is a side-effect to this, which is that round-robin will be a bit
unbalanced as the data nodes that are also selected as proxies will
receive more requests.

Cross-cluster search selects a subset of nodes for each remote cluster and sends requests only to them, which will act as a proxy and properly redirect such requests to the target nodes that hold the relevant data. What happens today is that every time we send a request to a remote cluster, it will be sent to the next node in the proxy list (in round-robin fashion), regardless of whether the target node is already amongst the ones that we are connected to. In case for instance we need to send a shard search request to a data node that's also one of the selected proxy nodes, we may end up sending the request to it through one of the other proxy nodes. This commit optimizes this case to make sure that whenever we are already connected to a remote node, we will send a direct request rather than using the next proxy node. There is a side-effect to this, which is that round-robin will be a bit unbalanced as the data nodes that are also selected as proxies will receive more requests.

elasticmachine · 2018-06-12T13:41:15Z

Pinging @elastic/es-search-aggs

s1monw

left some suggestions LGTM

s1monw · 2018-06-12T14:54:22Z

server/src/main/java/org/elasticsearch/transport/RemoteClusterConnection.java

     */
    Transport.Connection getConnection(DiscoveryNode remoteClusterNode) {
-        DiscoveryNode discoveryNode = connectedNodes.get();
+        if (connectedNodes.contains(remoteClusterNode)) {


should we use transportService .nodeConnected instead. I mean if we for instance have the local cluster configured as a remote cluster we get this optimization as well?

makes sense, great catch!

s1monw · 2018-06-12T14:54:30Z

server/src/main/java/org/elasticsearch/transport/RemoteClusterConnection.java


-            @Override
-            public void sendRequest(long requestId, String action, TransportRequest request, TransportRequestOptions options)
+    static class ProxyConnection implements Transport.Connection {


make the class final?

s1monw · 2018-06-12T14:54:43Z

server/src/main/java/org/elasticsearch/transport/RemoteClusterConnection.java

    }

-    private static class ConnectedNodes implements Supplier<DiscoveryNode> {
+    private static class ConnectedNodes {


also final?

* elastic/master: (40 commits) [DOC] Extend SQL docs Immediately flush channel after writing to buffer (elastic#31301) [DOCS] Shortens ML API intros Use quotes in the call invocation (elastic#31249) move security ingest processors to a sub ingest directory (elastic#31306) Add 5.6.11 version constant. Fix version detection. SQL: Whitelist SQL utility class for better scripting (elastic#30681) [Docs] All Rollup docs experimental, agg limitations, clarify DeleteJob (elastic#31299) CCS: don't proxy requests for already connected node (elastic#31273) Mute ScriptedMetricAggregatorTests testSelfReferencingAggStateAfterMap [test] opensuse packaging turn up debug logging Add unreleased version 6.3.1 Removes experimental tag from scripted_metric aggregation (elastic#31298) [Rollup] Metric config parser must use builder so validation runs (elastic#31159) [ML] Check licence when datafeeds use cross cluster search (elastic#31247) Add notion of internal index settings (elastic#31286) Test: Remove broken yml test feature (elastic#31255) REST hl client: cluster health to default to cluster level (elastic#31268) [ML] Update test thresholds to account for changes to memory control (elastic#31289) ...

* elastic/master: (29 commits) [DOC] Extend SQL docs Immediately flush channel after writing to buffer (elastic#31301) [DOCS] Shortens ML API intros Use quotes in the call invocation (elastic#31249) move security ingest processors to a sub ingest directory (elastic#31306) Add 5.6.11 version constant. Fix version detection. SQL: Whitelist SQL utility class for better scripting (elastic#30681) [Docs] All Rollup docs experimental, agg limitations, clarify DeleteJob (elastic#31299) CCS: don't proxy requests for already connected node (elastic#31273) Mute ScriptedMetricAggregatorTests testSelfReferencingAggStateAfterMap [test] opensuse packaging turn up debug logging Add unreleased version 6.3.1 Removes experimental tag from scripted_metric aggregation (elastic#31298) [Rollup] Metric config parser must use builder so validation runs (elastic#31159) [ML] Check licence when datafeeds use cross cluster search (elastic#31247) Add notion of internal index settings (elastic#31286) Test: Remove broken yml test feature (elastic#31255) REST hl client: cluster health to default to cluster level (elastic#31268) [ML] Update test thresholds to account for changes to memory control (elastic#31289) ...

* master: Remove RestGetAllAliasesAction (#31308) Temporary fix for broken build Reenable Checkstyle's unused import rule (#31270) Remove remaining unused imports before merging #31270 Fix non-REST doc snippet [DOC] Extend SQL docs Immediately flush channel after writing to buffer (#31301) [DOCS] Shortens ML API intros Use quotes in the call invocation (#31249) move security ingest processors to a sub ingest directory (#31306) Add 5.6.11 version constant. Fix version detection. SQL: Whitelist SQL utility class for better scripting (#30681) [Docs] All Rollup docs experimental, agg limitations, clarify DeleteJob (#31299) CCS: don't proxy requests for already connected node (#31273) Mute ScriptedMetricAggregatorTests testSelfReferencingAggStateAfterMap [test] opensuse packaging turn up debug logging Add unreleased version 6.3.1 Removes experimental tag from scripted_metric aggregation (#31298) [Rollup] Metric config parser must use builder so validation runs (#31159) [ML] Check licence when datafeeds use cross cluster search (#31247) Add notion of internal index settings (#31286) Test: Remove broken yml test feature (#31255) REST hl client: cluster health to default to cluster level (#31268) [ML] Update test thresholds to account for changes to memory control (#31289) Log warnings when cluster state publication failed to some nodes (#31233) Fix AntFixture waiting condition (#31272) Ignore numeric shard count if waiting for ALL (#31265) [ML] Implement new rules design (#31110) index_prefixes back-compat should test 6.3 (#30951) Core: Remove plain execute method on TransportAction (#30998) Update checkstyle to 8.10.1 (#31269) Set analyzer version in PreBuiltAnalyzerProviderFactory (#31202) Modify pipelining handlers to require full requests (#31280) Revert upgrade to Netty 4.1.25.Final (#31282) Use armored input stream for reading public key (#31229) Fix Netty 4 Server Transport tests. Again. REST hl client: adjust wait_for_active_shards param in cluster health (#31266) REST high-level Client: remove deprecated API methods (#31200) [DOCS] Mark SQL feature as experimental [DOCS] Updates machine learning custom URL screenshots (#31222) Fix naming conventions check for XPackTestCase Fix security Netty 4 transport tests Fix race in clear scroll (#31259) [DOCS] Clarify audit index settings when remote indexing (#30923) Delete typos in SAML docs (#31199) REST high-level client: add Cluster Health API (#29331) [ML][TEST] Mute tests using rules (#31204) Support RequestedAuthnContext (#31238) SyncedFlushResponse to implement ToXContentObject (#31155) Add Get Aliases API to the high-level REST client (#28799) Remove some line length supressions (#31209) Validate xContentType in PutWatchRequest. (#31088) [INGEST] Interrupt the current thread if evaluation grok expressions take too long (#31024) Suppress extras FS on caching directory tests Revert "[DOCS] Added 6.3 info & updated the upgrade table. (#30940)" Revert "Fix snippets in upgrade docs" Fix snippets in upgrade docs [DOCS] Added 6.3 info & updated the upgrade table. (#30940) LLClient: Support host selection (#30523) Upgrade to Netty 4.1.25.Final (#31232) Enable custom credentials for core REST tests (#31235) Move ESIndexLevelReplicationTestCase to test framework (#31243) Encapsulate Translog in Engine (#31220) HLRest: Add get index templates API (#31161) Remove all unused imports and fix CRLF (#31207) [Tests] Fix self-referencing tests [TEST] Fix testRecoveryAfterPrimaryPromotion [Docs] Remove mention pattern files in Grok processor (#31170) Use stronger write-once semantics for Azure repository (#30437) Don't swallow exceptions on replication (#31179) Limit the number of concurrent requests per node (#31206) Call ensureNoSelfReferences() on _agg state variable after scripted metric agg script executions (#31044) Move java version checker back to its own jar (#30708) [test] add fix for rare virtualbox error (#31212)

Cross-cluster search selects a subset of nodes for each remote cluster and sends requests only to them, which will act as a proxy and properly redirect such requests to the target nodes that hold the relevant data. What happens today is that every time we send a request to a remote cluster, it will be sent to the next node in the proxy list (in round-robin fashion), regardless of whether the target node is already amongst the ones that we are connected to. In case for instance we need to send a shard search request to a data node that's also one of the selected proxy nodes, we may end up sending the request to it through one of the other proxy nodes. This commit optimizes this case to make sure that whenever we are already connected to a remote node, we will send a direct request rather than using the next proxy node. There is a side-effect to this, which is that round-robin will be a bit unbalanced as the data nodes that are also selected as proxies will receive more requests.

* 6.x: Upgrade to Lucene-7.4.0-snapshot-518d303506 (#31360) [ML] Implement new rules design (#31110) (#31294) Remove RestGetAllAliasesAction (#31308) CCS: don't proxy requests for already connected node (#31273) Rankeval: Fold template test project into main module (#31203) [Docs] Remove reference to repository-s3 plugin creating an S3 bucket (#31359) More detailed tracing when writing metadata (#31319) Add details section for dcg ranking metric (#31177)

javanna added 2 commits June 11, 2018 17:46

Merge branch 'master' into enhancement/ccs_direct_if_connected

43c0d5e

javanna added >enhancement :Search/Search Search-related issues that do not fall into other categories v7.0.0 v6.4.0 labels Jun 12, 2018

javanna requested a review from s1monw June 12, 2018 13:41

s1monw approved these changes Jun 12, 2018

View reviewed changes

javanna added 2 commits June 13, 2018 12:38

address review comments

8979a5b

Merge branch 'master' into enhancement/ccs_direct_if_connected

193285a

javanna merged commit 664903a into elastic:master Jun 13, 2018

javanna added the backport pending label Jun 13, 2018

javanna removed the backport pending label Jun 15, 2018

colings86 added v7.0.0-beta1 and removed v7.0.0 labels Feb 7, 2019

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

CCS: don't proxy requests for already connected node#31273

CCS: don't proxy requests for already connected node#31273
javanna merged 4 commits intoelastic:masterfrom
javanna:enhancement/ccs_direct_if_connected

javanna commented Jun 12, 2018

Uh oh!

elasticmachine commented Jun 12, 2018

Uh oh!

s1monw left a comment

Uh oh!

s1monw Jun 12, 2018

Uh oh!

javanna Jun 13, 2018

Uh oh!

s1monw Jun 12, 2018

Uh oh!

s1monw Jun 12, 2018

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants

Conversation

javanna commented Jun 12, 2018

Uh oh!

elasticmachine commented Jun 12, 2018

Uh oh!

s1monw left a comment

Choose a reason for hiding this comment

Uh oh!

s1monw Jun 12, 2018

Choose a reason for hiding this comment

Uh oh!

javanna Jun 13, 2018

Choose a reason for hiding this comment

Uh oh!

s1monw Jun 12, 2018

Choose a reason for hiding this comment

Uh oh!

s1monw Jun 12, 2018

Choose a reason for hiding this comment

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants