[IMPALA-12043] Large catalog info triggers "TTransportException: MaxMessageSize reached"

• Type: Bug
• Status: Resolved
• Priority: Major
• Resolution: Fixed
• Affects Version/s: Impala 4.2.0
• Fix Version/s: Impala 4.3.0
• Component/s: Catalog
• Epic Color: ghx-label-2
• Assignee: Riza Suminto
• Reporter: Michael Smith
• Created: 05/Apr/23 21:13
• Updated: 11/Apr/23 12:35
• Resolved: 11/Apr/23 12:35

Description

Large results from something like COMPUTE INCREMENTAL STATS can result in a warning like

W0402 18:19:03.456277 250159 CatalogdMetaProvider.java:1235] 774b8a34875b106f:71548b8b00000000] Unable to deserialize updated catalog info. Skipping cache invalidation which may result in stale metadata being used at this coordinator.
Java exception follows:
org.apache.thrift.transport.TTransportException: MaxMessageSize reached
	at org.apache.thrift.transport.TEndpointTransport.resetConsumedMessageSize(TEndpointTransport.java:58)

This happens when https://github.com/apache/impala/blob/master/fe/src/main/java/org/apache/impala/catalog/local/CatalogdMetaProvider.java#L1231 receives a buffer through the JNI from NativeGetNextCatalogObjectUpdate that exceeds 100MB. TByteBuffer inherits from TEndpointTransport, which in Thrift 0.16.0 adds a MaxMessageSize limit. TMemoryBuffer adds a constructor that allows passing in a TConfiguration object to customize the limit, which we make use of in IMPALA-11669, but TByteBuffer does not provide a similar interface (THRIFT-5696) and was overlooked.

Issue Links

• depends upon: THRIFT-5696 TByteBuffer.java does not allow non-default TConfiguration
• is caused by: IMPALA-11385 Upgrade JAVA thrift components to thrift-0.16.0
• relates to: IMPALA-11669 Make Thrift max message size configuration