Listening to multiple tables causes server to crash

[jjbrosnan]

Description

The merged listener crashes the server when listening to multiple tables.

Steps to reproduce

from deephaven.table_listener import merged_listen
from deephaven import time_table

t1 = time_table("PT2s").update(["X = i"])
t2 = time_table("PT3s").update(["Y = ii"])

def listener_function(updates, is_replay):
    data_added_t1 = updates[t1].added()
    data_added_t2 = updates[t2].added()
    print(data_added_t1)
    print(data_added_t2)

handle = merged_listen([t1, t2], listener_function)

Expected results

One of two things:

• The code to work and print data added to each table when they tick.
• An error message giving me some indication of what I didn't do right.

Actual results

The merged listener appears to run for a second or two, printing stuff. Then, the server crashes, and this is displayed:

Server shutdown: Exception while processing UpdateGraph notification

java.lang.NullPointerException: Cannot invoke "io.deephaven.engine.table.impl.QueryTable.satisfied(long)" because "this.result" is null
io.deephaven.engine.table.impl.MergedListener.propagateErrorDownstream(MergedListener.java:175)
io.deephaven.engine.table.impl.MergedListener.propagateError(MergedListener.java:154)
io.deephaven.engine.table.impl.MergedListener.handleUncaughtException(MergedListener.java:300)
io.deephaven.engine.table.impl.MergedListener$MergedNotification.run(MergedListener.java:368)
io.deephaven.engine.updategraph.impl.BaseUpdateGraph.runNotification(BaseUpdateGraph.java:711)
io.deephaven.engine.updategraph.impl.PeriodicUpdateGraph$ConcurrentNotificationProcessor.processSatisfiedNotifications(PeriodicUpdateGraph.java:836)
io.deephaven.engine.updategraph.impl.PeriodicUpdateGraph$NotificationProcessorThreadFactory.lambda$newThread$0(PeriodicUpdateGraph.java:1109)
java.base/java.lang.Thread.run(Thread.java:1583)

Additional details and attachments

If applicable, add any additional screenshots, logs, or other attachments to help explain your problem.

Versions

• Deephaven: edge
• OS: OS X
• Browser: Chrome
• Docker: 20.10.13

[jjbrosnan]

Note that adding a try/except block in the listener function itself resolves the crash.

[rcaudy]

In general, you aren't supposed to throw uncaught exceptions in listeners. We probably need to wrap the listener callback to address that, but the issue is that we don't actually have a good way to signal that to the user. Maybe we need a secondary callback for "my listener blew up"?