[IGNITE-18410] Put data to metastorage during some components start leads to assertion error

• Type: Bug
• Status: Resolved
• Priority: Major
• Resolution: Fixed
• Fix Version/s: 3.0
• Labels: ignite-3
• Ignite Flags: Docs Required, Release Notes Required
• Assignee: Ivan Bessonov
• Reporter: Mirza Aliev
• Reviewer: Mirza Aliev
• Created: 14/Dec/22 08:34
• Updated: 04/Sep/24 16:19
• Resolved: 19/Dec/22 13:49

Description

When we was developing https://issues.apache.org/jira/browse/IGNITE-18087 some tests on restart started to fail on the assertion

assert rev >= appliedRevision : IgniteStringFormatter.format(

in ConfigurationCatchUpListener.

It means that we have inconsistency when we compare storage revision from configuration and revision from the event. Needed to investigate.

UPD:

We made a fix in DistributedConfigurationStorage#resolveRevision and added else section were we initialised master key with 0 when there was no changes in configuration, but some puts to the metastorage were done, so the DistributedConfigurationStorage#resolveRevision method returned wrong non-null revision for the configuration. But seems that this is not enough and we still get some failed tests.

We fail we the assertion

assert data.isEmpty() || cfgRevision > 0;

in DistributedConfigurationStorage#readDataOnRecovery0 (see muted test ItDistributedConfigurationStorageTest#testRestartWithPds)

or get exception

                    if (revision <= appliedRevision) {
                        throw new MetaStorageException(DEPLOYING_WATCH_ERR, String.format(
                                "Current revision (%d) must be greater than the revision in the Vault (%d)",
                                revision, appliedRevision
                        ));
                    }

in

MetaStorageManager#storeEntries (see muted test ItTablesApiTest#testGetTableFromLaggedNode)

Issue Links

• blocks: IGNITE-18087 Populate DistributionZoneManager with CMG listeners to logical topology events
• is related to: IGNITE-18429 ItTablesApiTest#testTableAlreadyCreatedFromLaggedNode is flaky
• links to: GitHub Pull Request #1455