[STORM-3687] Add a warning about possible issues on a mixed cluster if the StormCommon.systemTopology implementation is changed

• Type: Comment
• Status: Resolved
• Priority: Minor
• Resolution: Fixed
• Fix Version/s: 2.3.0
• Assignee: Ethan Li
• Reporter: Ethan Li
• Created: 05/Aug/20 16:27
• Updated: 06/Aug/20 04:37
• Resolved: 06/Aug/20 04:36

Description

During rolling upgrade, some supervisors are still at 2.2, and others are running on 2.3. So some workers are running with storm-client-2.2.jar and others are on 2.3.

Because of https://github.com/apache/storm/commit/93a7f770d508668bc7af183e08535813fff6f805 (STORM-3660)

the "_credentials" stream is removed.
So the map of streamId to streamName for system component is changed
refer to the code at:
https://github.com/apache/storm/blob/v2.2.0/storm-client/src/jvm/org/apache/storm/serialization/SerializationFactory.java#L218-L222

So a worker running on storm-client-2.2 might send out _metrics from its systembolt, while the MetricsConsumer running with storm-client-2.3 interprets it as "_metrics_tick" because the mapping is different:

o.a.s.s.SerializationFactory Thread-15-__system-executor[-1, -1] [INFO] idmap for system comp {__tick=6, __system=5, __metrics_tick=4, __credentials=1, __flush=2, __metrics=3}

vs

o.a.s.s.SerializationFactory Netty-server-localhost-6703-worker-1 [INFO] idmap for system comp {__tick=5, __system=4, __metrics_tick=3, __flush=1, __metrics=2}

Hence we see

Caused by: java.lang.RuntimeException: java.lang.ClassCastException: org.apache.storm.metric.api.IMetricsConsumer$TaskInfo cannot be cast to java.lang.Integer
        at org.apache.storm.executor.Executor.accept(Executor.java:293) ~[storm-client-2.3.0.y.jar:2.3.0.y]
        at org.apache.storm.utils.JCQueue.consumeImpl(JCQueue.java:113) ~[storm-client-2.3.0.y.jar:2.3.0.y]
        at org.apache.storm.utils.JCQueue.consume(JCQueue.java:89) ~[storm-client-2.3.0.y.jar:2.3.0.y]
        at org.apache.storm.executor.bolt.BoltExecutor$1.call(BoltExecutor.java:167) ~[storm-client-2.3.0.y.jar:2.3.0.y]
        at org.apache.storm.executor.bolt.BoltExecutor$1.call(BoltExecutor.java:153) ~[storm-client-2.3.0.y.jar:2.3.0.y]
        at org.apache.storm.utils.Utils$1.run(Utils.java:398) ~[storm-client-2.3.0.y.jar:2.3.0.y]
        ... 1 more
Caused by: java.lang.ClassCastException: org.apache.storm.metric.api.IMetricsConsumer$TaskInfo cannot be cast to java.lang.Integer
        at org.apache.storm.tuple.TupleImpl.getInteger(TupleImpl.java:121) ~[storm-client-2.3.0.y.jar:2.3.0.y]
        at org.apache.storm.executor.Executor.metricsTick(Executor.java:320) ~[storm-client-2.3.0.y.jar:2.3.0.y]
        at org.apache.storm.executor.bolt.BoltExecutor.tupleActionFn(BoltExecutor.java:213) ~[storm-client-2.3.0.y.jar:2.3.0.y]
        at org.apache.storm.executor.Executor.accept(Executor.java:286) ~[storm-client-2.3.0.y.jar:2.3.0.y]
        at org.apache.storm.utils.JCQueue.consumeImpl(JCQueue.java:113) ~[storm-client-2.3.0.y.jar:2.3.0.y]
        at org.apache.storm.utils.JCQueue.consume(JCQueue.java:89) ~[storm-client-2.3.0.y.jar:2.3.0.y]
        at org.apache.storm.executor.bolt.BoltExecutor$1.call(BoltExecutor.java:167) ~[storm-client-2.3.0.y.jar:2.3.0.y]
        at org.apache.storm.executor.bolt.BoltExecutor$1.call(BoltExecutor.java:153) ~[storm-client-2.3.0.y.jar:2.3.0.y]
        at org.apache.storm.utils.Utils$1.run(Utils.java:398) ~[storm-client-2.3.0.y.jar:2.3.0.y]
        ... 1 more
2020-08-04 18:26:08.017 o.a.s.m.c



2020-08-04 18:26:08.027 o.a.s.u.Utils Thread-19-__metrics_org.apache.storm.metric.LoggingMetricsConsumer-executor[1555, 1555] [ERROR] Halting process: Worker died
java.lang.RuntimeException: Halting process: Worker died
        at org.apache.storm.utils.Utils.exitProcess(Utils.java:518) [storm-client-2.3.0.y.jar:2.3.0.y]
        at org.apache.storm.utils.Utils$3.run(Utils.java:870) [storm-client-2.3.0.y.jar:2.3.0.y]
        at org.apache.storm.executor.error.ReportErrorAndDie.uncaughtException(ReportErrorAndDie.java:41) [storm-client-2.3.0.y.jar:2.3.0.y]
        at java.lang.Thread.dispatchUncaughtException(Thread.java:1959) [?:1.8.0_242]
2020-08-04 18:26:08.027 o.a.s.u.Utils Thread-18-__metrics_org.apache.storm.metric.LoggingMetricsConsumer-executor[1553, 1553] [ERROR] Halting process: Worker died
java.lang.RuntimeException: Halting process: Worker died
        at org.apache.storm.utils.Utils.exitProcess(Utils.java:518) [storm-client-2.3.0.y.jar:2.3.0.y]
        at org.apache.storm.utils.Utils$3.run(Utils.java:870) [storm-client-2.3.0.y.jar:2.3.0.y]
        at org.apache.storm.executor.error.ReportErrorAndDie.uncaughtException(ReportErrorAndDie.java:41) [storm-client-2.3.0.y.jar:2.3.0.y]
        at java.lang.Thread.dispatchUncaughtException(Thread.java:1959) [?:1.8.0_242]
2020-08-04 18:26:08.028 o.a.s.u.Utils ShutdownHook-sleepKill-3s [INFO] Halting after 3 seconds

Since mixed cluster is not guaranteed to work, I am not going to fix this for now. But will add some warnings in the code so people can be aware of it.

Issue Links

• relates to: STORM-3660 Remove messaging queue for updating Executor credentials