SAI-4837 : In-memory synthetic `SolrCore` for QA node #181

patsonluk · 2024-02-24T04:35:44Z

Description

A prototype to attempt using an in-memory SyntheticSolrCore similar to the SolrCoreProxy in our Solr 8.8

The synthetic collection/replica introduced in Solr 9 added the collection folder to physically disk and also the zookeeper solr/collections.

However, it seems that what we essentially need is a SolrCore in the memory for each config set, which might be able to simply register it against the CoreContainer without actually registering to ZK.

Solution

Introduced SyntheticSolrCore which extends SolrCore, SyntheticSolrCore is only used in Coordinator node to support a subset of SolrCore functionalities required by Coordinator operations such as aggregating and writing out response and providing configset info.

Such core is only registered to the CoreContainer but is no longer register itself to zookeeper.

Take note that to minimize code changes, there will still be snapshot/index data directory written to the QA node. We could have overridden initSnapshotMetaDataManager and initIndex to be no-op to avoid creating them, however, there is currently other code (and possibly future code too) that might rely on the existence of those values. It is safer to just keep them as is.

The data folder for the synthetic core would NOT be loaded on QA node start-up as there's no core.properties in such directory. This is because instead of going through CoreContainer#create for such synthetic core, it's using SyntheticSolrCore#createAndRegisterCore, which bypass core.properties creation and zk registration

Tests

TestCoordinatorRole passed, which provides good integration test coverage to ensure SyntheticSolrCore is working as expected
Local and playpen testing:
i. Basic query and restart tests from QA node
ii. Shard split and moves and ensure QA node can distribute the requests to new shards
iii. collection name is logged correctly
iv. test on multiple collections to ensure that query work for them all, and the synthetic core should only be created once (by observing the log message Loading synthetic core for config set...)

hiteshk25 · 2024-02-26T17:44:19Z

solr/core/src/java/org/apache/solr/core/SolrCore.java

@@ -1175,7 +1178,7 @@ private SolrCore(
      initSearcher(prev);

      // Initialize the RestManager
-      restManager = initRestManager();
+      restManager = isSynthetic() ? new RestManager() : initRestManager();


how this is helping here?

It was a quick prototype so I basically quickly handle NPEs when i ran into them. 😁 I put this one is so it will init w/o throwing exceptions.

I believe for a better implementation, we will need to understand all the init happened in the SolrCore, there's a fair chance that we can simply bypass most of the init routine for a synthetic core (rest manager, event handler/listener, plugin, searcher etc), as synthetic core does not do any real searching, it simply relays it to the actual data node

hiteshk25 · 2024-02-26T17:45:01Z

solr/core/src/java/org/apache/solr/cloud/ZkConfigSetService.java

@@ -66,6 +66,12 @@ public ZkConfigSetService(SolrZkClient zkClient) {
    this.zkClient = zkClient;
  }

+  @Override
+  protected SolrResourceLoader createConfigSetResourceLoader(CoreDescriptor cd, String configSetName) {


to understand better, why we need to override here?

For synthetic core, the resource loader only needs to load the configset resource, it would have no access to collection specific resource - which is non existent for this new synthetic core/collection, since it does not even exists in ZK.

There are several implementation of ConfigSetService, and for prototype purpose, we only support this for ZkConfigSetSrvice

hiteshk25 · 2024-02-26T17:46:30Z

solr/core/src/java/org/apache/solr/servlet/CoordinatorHttpSolrCall.java

-                e);
-          }
-        }
+        coreContainer.registerCore(syntheticCoreDescriptor, syntheticCore, false, false);


not publishing into zk? nice!

patsonluk · 2024-02-26T18:03:55Z

Also pushed some fixes, so the unit test case would pass

patsonluk · 2024-04-17T22:06:47Z

@chatman @noblepaul . I would ❤️ to gather some thoughts from you both! Mostly on the idea, as for the implementation details, we might need to further cleanup the logic if we decide to move forward.

The synthetic collection was created to avoid excess proxy cores created in 8.x AFAIK, I'm wondering if this proposal could address such concern without raising new ones?

noblepaul · 2024-04-17T22:25:20Z

The biggest challenge is pushing this upstream. Making changes to SolrCore and introducing a concept of synthetic for a one of the core classes, for a niche feature is hardsell. However if this is to be only kept in-house , this should be fine

patsonluk · 2024-04-18T01:32:26Z

The biggest challenge is pushing this upstream. Making changes to SolrCore and introducing a concept of synthetic for a one of the core classes, for a niche feature is hardsell. However if this is to be only kept in-house , this should be fine

That's a good point, totally agree! Even if we do not upstream this, minimizing changes to SolrCore could save us from merge issues.

I have added some minor changes and javadoc to my implementation. The change to SolrCore should now only be the access modifier of bufferUpdatesIfConstructing. I'm hoping that the apache community will find it not as invasive (and it used to be protected as well in 8.8).

noblepaul · 2024-04-22T07:30:49Z

I like the fact that the changes to SolrCore is trivial/non-invasive. We can push this upstream as well

👍

noblepaul

Looks good

hiteshk25 · 2024-04-22T17:04:43Z

solr/core/src/java/org/apache/solr/core/SyntheticSolrCore.java

+    // core's collection from ZK
+    // which synthetic core is not registered in ZK.
+    // We do not expect RestManager ops on Coordinator Nodes
+    return new RestManager();


what restManager enables?

Actually, there is only a tiny feature called "managed resources" being powered by RestManager. It's not even reuired

hiteshk25

LGTM

(Thanks @patsonluk , please make sure QA logs are ok (collname, etc), though I don't feel any change in that area)

Ensure only one synthetic core created per config set

patsonluk · 2024-04-22T22:58:46Z

@hiteshk25

Good call on please make sure QA logs are ok (collname, etc) . The existing proposal did miss the collection name!

I have committed some fixes:

Collection name should be set properly for the collection being queried on now.
The SyntheticSolrCore should now be one instance per configset, my previous code didn't check correctly

Would you mind to review the changes in a32a2ac . Many thanks again!!!

hiteshk25

LGTM (Thanks @patsonluk )

Should open the registered core on first synthetic core creation since it is a getCore op

* Experimental QA core as proxy in memory * Experimental QA core as proxy in memory * Experimental QA core as proxy in memory * ./gradlew tidy * Comment out portion that checks for synthetic collection/core existence * Fixes watches and test cases * Refactoring to avoid modifying existing method modifiers * Cleanup * Some refactoring to minimize changes to SolrCore * javadoc * ./gradlew tidy * code cleanup * code cleanup * Fix resource loader issue if configSetName is null * ./gradlew tidy * Fixed issue with incorrect collection name in context Ensure only one synthetic core created per config set * ./gradlew tidy * Correctly use <config set>_core as synthetic core name for clarity Should open the registered core on first synthetic core creation since it is a getCore op * ./gradlew tidy * Fixed issue with incorrect collection name in context

hiteshk25 reviewed Feb 26, 2024

View reviewed changes

patsonluk added 9 commits April 17, 2024 17:46

Experimental QA core as proxy in memory

7e2ffb6

Experimental QA core as proxy in memory

f6ce0c1

Experimental QA core as proxy in memory

64b5293

./gradlew tidy

49cfa7b

Comment out portion that checks for synthetic collection/core existence

5de2a27

Fixes watches and test cases

924391a

Refactoring to avoid modifying existing method modifiers

6772c22

Cleanup

fa08d7a

Some refactoring to minimize changes to SolrCore

0d2b744

patsonluk force-pushed the patsonluk/qa-core-proxy branch from 5b7b8c1 to 0d2b744 Compare April 18, 2024 00:47

javadoc

a1582f6

./gradlew tidy

bff8168

patsonluk changed the title ~~Experiment : in-memory Solr Core Proxy for QA node~~ Experiment : in-memory synthetic SolrCore for QA node Apr 18, 2024

code cleanup

28e5669

patsonluk marked this pull request as ready for review April 18, 2024 21:36

code cleanup

dd259c0

patsonluk changed the title ~~Experiment : in-memory synthetic SolrCore for QA node~~ In-memory synthetic SolrCore for QA node Apr 18, 2024

patsonluk added 2 commits April 18, 2024 16:27

Fix resource loader issue if configSetName is null

b8a15e7

./gradlew tidy

bb11583

patsonluk assigned noblepaul, chatman and hiteshk25 Apr 19, 2024

noblepaul approved these changes Apr 22, 2024

View reviewed changes

hiteshk25 reviewed Apr 22, 2024

View reviewed changes

hiteshk25 approved these changes Apr 22, 2024

View reviewed changes

patsonluk added 2 commits April 22, 2024 15:44

Fixed issue with incorrect collection name in context

a32a2ac

Ensure only one synthetic core created per config set

./gradlew tidy

02a560f

hiteshk25 approved these changes Apr 22, 2024

View reviewed changes

patsonluk added 3 commits April 22, 2024 18:02

Correctly use <config set>_core as synthetic core name for clarity

ef0ed24

Should open the registered core on first synthetic core creation since it is a getCore op

./gradlew tidy

46cfa39

Fixed issue with incorrect collection name in context

8052d83

patsonluk merged commit cba3230 into fs/branch_9_3 May 2, 2024
7 checks passed

patsonluk deleted the patsonluk/qa-core-proxy branch May 2, 2024 21:05

patsonluk changed the title ~~In-memory synthetic SolrCore for QA node~~ SAI-4837 : In-memory synthetic SolrCore for QA node May 2, 2024

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

SAI-4837 : In-memory synthetic `SolrCore` for QA node #181

SAI-4837 : In-memory synthetic `SolrCore` for QA node #181

patsonluk commented Feb 24, 2024 •

edited

Loading

hiteshk25 Feb 26, 2024

patsonluk Feb 26, 2024 •

edited

Loading

hiteshk25 Feb 26, 2024

patsonluk Feb 26, 2024

hiteshk25 Feb 26, 2024

patsonluk commented Feb 26, 2024

patsonluk commented Apr 17, 2024

noblepaul commented Apr 17, 2024

patsonluk commented Apr 18, 2024 •

edited

Loading

noblepaul commented Apr 22, 2024 •

edited

Loading

noblepaul left a comment

hiteshk25 Apr 22, 2024

noblepaul Apr 22, 2024

hiteshk25 left a comment

patsonluk commented Apr 22, 2024

hiteshk25 left a comment

SAI-4837 : In-memory synthetic SolrCore for QA node #181

SAI-4837 : In-memory synthetic SolrCore for QA node #181

Conversation

patsonluk commented Feb 24, 2024 • edited Loading

Description

Solution

Tests

hiteshk25 Feb 26, 2024

Choose a reason for hiding this comment

patsonluk Feb 26, 2024 • edited Loading

Choose a reason for hiding this comment

hiteshk25 Feb 26, 2024

Choose a reason for hiding this comment

patsonluk Feb 26, 2024

Choose a reason for hiding this comment

hiteshk25 Feb 26, 2024

Choose a reason for hiding this comment

patsonluk commented Feb 26, 2024

patsonluk commented Apr 17, 2024

noblepaul commented Apr 17, 2024

patsonluk commented Apr 18, 2024 • edited Loading

noblepaul commented Apr 22, 2024 • edited Loading

noblepaul left a comment

Choose a reason for hiding this comment

hiteshk25 Apr 22, 2024

Choose a reason for hiding this comment

noblepaul Apr 22, 2024

Choose a reason for hiding this comment

hiteshk25 left a comment

Choose a reason for hiding this comment

patsonluk commented Apr 22, 2024

hiteshk25 left a comment

Choose a reason for hiding this comment

SAI-4837 : In-memory synthetic `SolrCore` for QA node #181

SAI-4837 : In-memory synthetic `SolrCore` for QA node #181

patsonluk commented Feb 24, 2024 •

edited

Loading

patsonluk Feb 26, 2024 •

edited

Loading

patsonluk commented Apr 18, 2024 •

edited

Loading

noblepaul commented Apr 22, 2024 •

edited

Loading