CREATE TABLE leaves /tmp/presto-{USER}/xxx

[byungnam]

Problem

While using Presto, I found 1 million of /tmp/presto-{USER}/xxx directories and my Hadoop namenode is complaining about it: The directory item limit of /tmp/presto-**** is exceeded.

It seems something went wrong and the directory is not cleaned up.

Presto version

331

How to reproduce

Actually, no idea.

Not sure but we have many Presto query failures in some circumstances including when upgrading and restarting Presto, Hive fails to write HDFS (due to heavy IO), Presto worker fails (OOM is observed). So that could be a reason maybe?

ref: Slack thread link

[byungnam]

I found the cause of this behavior. It's still happening from Trino 355.

EDIT: This was not the cause. See the following comment.

How to reproduce:

1. Call call system.create_empty_partition()
2. Above call fails (due to partition already exists)

Example:
Call this

call system.create_empty_partition(schema_name => 'toros_analysis', table_name => 'o_aurochs_feedback_log', partition_columns => ARRAY['date_hour'], partition_values => ARRAY['2021091610'])

and fails with

io.trino.spi.TrinoException: Partition already exists
	at io.trino.plugin.hive.procedure.CreateEmptyPartitionProcedure.doCreateEmptyPartition(CreateEmptyPartitionProcedure.java:125)
	at io.trino.plugin.hive.procedure.CreateEmptyPartitionProcedure.createEmptyPartition(CreateEmptyPartitionProcedure.java:102)
	at java.base/java.lang.invoke.MethodHandle.invokeWithArguments(MethodHandle.java:710)
	at java.base/java.lang.invoke.MethodHandle.invokeWithArguments(MethodHandle.java:735)
	at io.trino.execution.CallTask.execute(CallTask.java:194)
	at io.trino.execution.CallTask.execute(CallTask.java:67)
	at io.trino.execution.DataDefinitionExecution.start(DataDefinitionExecution.java:170)
	at io.trino.execution.SqlQueryManager.createQuery(SqlQueryManager.java:237)
	at io.trino.dispatcher.LocalDispatchQuery.lambda$startExecution$7(LocalDispatchQuery.java:143)
	at io.trino.$gen.Trino_355_kakao____20210906_064002_2.run(Unknown Source)
	at java.base/java.util.concurrent.ThreadPoolExecutor.runWorker(ThreadPoolExecutor.java:1128)
	at java.base/java.util.concurrent.ThreadPoolExecutor$Worker.run(ThreadPoolExecutor.java:628)
	at java.base/java.lang.Thread.run(Thread.java:834)

then the /tmp/presto-{USER}/xxx directory is created but not removed.

drwxrwxrwx   - gony-noreply hadoop          0 2021-09-16 10:20 /tmp/presto-gony-noreply/5f4a14e7-3785-4036-aa1d-3fb18705e20b

[byungnam]

EDIT: It's CREATE TABLE statement not create_empty_partition

Reproduce steps:

• Run CREATE TABLE statement

[byungnam]

fixed by: #14830