[FLUME-3352] Unnecessary canary test will block on readonly spooldir while another trackerdir is set.

• Type: Bug
• Status: Open
• Priority: Blocker
• Resolution: Unresolved
• Affects Version/s: 1.9.0
• Component/s: Sinks+Sources
• Labels: patch pull-request-available
• Flags: Patch, Important
• Assignee: Unassigned
• Reporter: taoyang
• Created: 16/Jan/20 11:56
• Updated: 01/Jun/22 06:25

Description

Phenomenon

In many cases, we have just read permission on spoolDir and write permission on trackerDir.

However whenever flume starts for spooldir source , it will always try to create a '.canary' file in the spooling directory.

Then it leads to the failure of some processing unnecessarily.

Recur

(Usually spooldir is mounted readonly from a nas in production and this time we create a readonly directory instead.)

First we create the spoolDir by root user and create a file in it and it is readonly for others by default.

su root
mkdir /home/hadoop/testspooldir
echo 'foo' > /home/hadoop/testspooldir/bar

Then switch to another user (hadoop) who runs flume and make sure it has read permission for the spooldir.

su hadoop
mkdir /home/hadoop/testtrackerdir
ll /home/hadoop/testspooldir
>> total 4-rw-r--r-- 1 root root 4 Jan 16 19:15 bar

now create the example.conf:

a1.sources = r1
a1.sinks = k1
a1.channels = c1
a1.sources.r1.type = spooldir
a1.sources.r1.deletePolicy = never
a1.sources.r1.spoolDir = /home/hadoop/testspooldir
a1.sources.r1.trackerDir = /home/hadoop/testtrackerdir
a1.sources.r1.trackingPolicy = tracker_dir
a1.sinks.k1.type = logger
a1.channels.c1.type = memory
a1.channels.c1.capacity = 1000
a1.channels.c1.transactionCapacity = 100
a1.sources.r1.channels = c1
a1.sinks.k1.channel = c1

and start flume with it

bin/flume-ng agent --conf conf --conf-file conf/example.conf --name a1 -Dflume.root.logger=INFO,console

then the IOException is thrown.

2020-01-16 19:16:12,777 (lifecycleSupervisor-1-0) [ERROR - org.apache.flume.lifecycle.LifecycleSupervisor$MonitorRunnable.run(LifecycleSupervisor.java:251)] Unabl2020-01-16 19:16:12,777 (lifecycleSupervisor-1-0) [ERROR - org.apache.flume.lifecycle.LifecycleSupervisor$MonitorRunnable.run(LifecycleSupervisor.java:251)] Unable to start EventDrivenSourceRunner: { source:Spool Directory source r1: { spoolDir: /home/hadoop/testspooldir } } - Exception follows.org.apache.flume.FlumeException: Unable to read and modify files in the spooling directory: /home/hadoop/testspooldir at org.apache.flume.client.avro.ReliableSpoolingFileEventReader.<init>(ReliableSpoolingFileEventReader.java:195) at org.apache.flume.client.avro.ReliableSpoolingFileEventReader.<init>(ReliableSpoolingFileEventReader.java:89) at org.apache.flume.client.avro.ReliableSpoolingFileEventReader$Builder.build(ReliableSpoolingFileEventReader.java:882) at org.apache.flume.source.SpoolDirectorySource.start(SpoolDirectorySource.java:111) at org.apache.flume.source.EventDrivenSourceRunner.start(EventDrivenSourceRunner.java:44) at org.apache.flume.lifecycle.LifecycleSupervisor$MonitorRunnable.run(LifecycleSupervisor.java:249) at java.util.concurrent.Executors$RunnableAdapter.call(Executors.java:511) at java.util.concurrent.FutureTask.runAndReset(FutureTask.java:308) at java.util.concurrent.ScheduledThreadPoolExecutor$ScheduledFutureTask.access$301(ScheduledThreadPoolExecutor.java:180) at java.util.concurrent.ScheduledThreadPoolExecutor$ScheduledFutureTask.run(ScheduledThreadPoolExecutor.java:294) at java.util.concurrent.ThreadPoolExecutor.runWorker(ThreadPoolExecutor.java:1149) at java.util.concurrent.ThreadPoolExecutor$Worker.run(ThreadPoolExecutor.java:624) at java.lang.Thread.run(Thread.java:748)Caused by: java.io.IOException: Permission denied at java.io.UnixFileSystem.createFileExclusively(Native Method) at java.io.File.createTempFile(File.java:2024) at org.apache.flume.client.avro.ReliableSpoolingFileEventReader.<init>(ReliableSpoolingFileEventReader.java:185) ... 12 more

Fix

We just add the condition where this trick is necessary.

The pr/patch will be submitted as as shown below.

Or let it still exist and using warning log instead of exception thrown is better?

Issue Links

• links to: GitHub Pull Request #314