Manish Nema created STORM-953:
---------------------------------
Summary: Acknowledging Unanchored tuple causes freeze in toplogy
Key: STORM-953
URL: https://issues.apache.org/jira/browse/STORM-953
Project: Apache Storm
Issue Type: Bug
Affects Versions: 0.9.3
Environment: RHEL 5.0
Oracle JDK 1.7
Reporter: Manish Nema
I am running a topology in the 6 node cluster on RHEL 5 and Oracle JDK 1.7.
Schematic for topology as attached, there are 6 workers for spout and bolts.
Topology picks up message from JMS and sends out message to bolts. Each bolt
does some processing and sends out next set of tuple to downstream for further
processing. All bolts are in shuffle grouping. There are two special bolts
1. Log
2. Ticket
Log bolt receive same tuple from each bolt for logging purpose but these tuples
are unanchored. Similarly Ticket bolt receive tuple that are failed due to
business logic, these are also unanchored. Bolt1 to Bolt4 and anchored tuple
since we wanted message guarantee for these tuples. Each bolt receive one tuple
and emit one data tuple from Bolt1 to Bolt4.
I have observed that we were acking some unanchored tuple in Log and Ticket
bolts, and it was causing topology freeze after sometime (10 mins when worker
jvm heap size is set to 2 GB , 2 days when worker jvm heap was set to 16 GB).
These freeze essentially slows down message read from JMS. After commenting out
acking I didnt observe any slow down/freeze.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
