[ATLAS-4155] NotificationHookConsumer: Large Compressed Message Processing Problem

• Type: Bug
• Status: Resolved
• Priority: Major
• Resolution: Fixed
• Affects Version/s: 2.1.0
• Component/s: atlas-core
• Labels: consumer hooks
• Assignee: Ashutosh Mestry
• Reporter: Ashutosh Mestry
• Created: 15/Feb/21 05:10
• Updated: 22/Feb/21 05:36
• Resolved: 22/Feb/21 05:36

Description

Background

Notification messages can be large in size. To get over Kafka's limitation on message size, Atlas has compressed and split messages. If message size goes beyond stipulated threshold, the message is compressed. If compressed message goes beyond the size, it is split into multiple messages.

Situation

Consider a message that is so large that uncompressing it takes longer than Kafka's timeout for message. This causes the problem where the large message offset is not committed in time and that causes Kafka to present the same message again.

Message Description:
Number of splits: 8
Compressed message size: 7,452,640
Uncompressed message size: 520,803,946
Time taken to uncompress and stitch messages: > 90 seconds

Sequence:
2021-02-10 14:57:24,221: first message received
2021-02-10 14:58:36,052: all splits combined – 72 seconds
2021-02-10 15:01:06,971: message processing completed – 90 seconds
2021-02-10 15:01:17,158: Kafka commit failed. Elapsed time since first message: 197 seconds
2021-02-10 15:01:19,857: attempt #2: first message received
2021-02-10 15:03:01,993: attempt #2: all splits combined – 102 seconds
2021-02-10 15:04:44,896: attempt #2: Kafka commit failed. Elapsed time since first message: 205 seconds
Back to #5

Solution

Maintain last offset received. If the same offset is presented, commit the offset and move on to the next message.

 

 

Issue Links

• links to: Code Review