[NIFI-13340] Flowfiles stopped to be ingested before a processor group

• Type: Bug
• Status: Resolved
• Priority: Major
• Resolution: Fixed
• Affects Version/s: 1.25.0
• Fix Version/s: 1.27.0, 2.0.0-M4
• Component/s: Core Framework
• Labels: data-consistency statestore
• Environment: Host OS :ubuntu 20.04 in wsl ; CPU: Intel(R) Xeon(R) Platinum 8370C CPU @ 2.80GHz ; RAM: 128GB ; ZFS ARC cache was configured to be maximum 4 GB
• Flags: Important
• Assignee: Mark Payne
• Reporter: Yuanhao Zhu
• Created: 03/Jun/24 13:11
• Updated: 27/Jun/24 14:48
• Resolved: 17/Jun/24 14:40

Description

Background: We run nifi as a standalone instance in a docker container in on all our stages, the statemanagement provider used by our instance is the local WAL backed provider.

Description: We observed that the flowfiles in front of one of our processor groups stopped to be ingested once in a while and it happens on all our stages without noticeable pattern. The flowfile concurrency policy of the processor group that stopped ingesting data is set to SINGLE_BATCH_PER_NODE and the outbound policy is set to BATCH_OUTPUT. The ingestion should continue since the processor group had already send out every flowfile in it, but it stopped. 

 

There is only one brutal solution from our side(We need to delete the processor group and restore it from nifi registry again) and the occurrence of this issue had impacted our data ingestion.

As you can see in the screenshot that the processor group 'Delete before Insert' has no more flowfile to output but still it does not ingest the data queued in the input port

 

In the log file I found the following:

 

2024-06-03 11:34:01,772 TRACE [Timer-Driven Process Thread-15] o.apache.nifi.groups.StandardDataValve Will not allow data to flow into StandardProcessGroup[identifier=5eb0ad69-e8ed-3ba0-52da-af94fb9836cd,name=Delete before Insert] because Outbound Policy is Batch Output and valve is already open to allow data to flow out of group

 

 
Also in the diagnostics, I found the following for the 'Delete before Insert' processor group:
 

Process Group e6510a87-aa78-3268-1b11-3c310f0ad144, Name = Search and Delete existing reports(This is the parent processor group of the Delete before Insert)
Currently Have Data Flowing In: []
Currently Have Data Flowing Out: [StandardProcessGroup[identifier=5eb0ad69-e8ed-3ba0-52da-af94fb9836cd,name=Delete before Insert]]
Reason for Not allowing data to flow in:
    Data Valve is already allowing data to flow out of group:
        StandardProcessGroup[identifier=5eb0ad69-e8ed-3ba0-52da-af94fb9836cd,name=Delete before Insert]
Reason for Not allowing data to flow out:
    Output is Allowed:
        StandardProcessGroup[identifier=5eb0ad69-e8ed-3ba0-52da-af94fb9836cd,name=Delete before Insert] 

 
Which is clearly not correct. since there are currently not data flowing out from the 'Delete before Insert' processor group.
 
We dig through the source code of StandardDataValve.java and found that that data valve's states are stored every time the data valve is opened and close so the potential reason causing this issue is that the processor group id was put into the statemap when data flowed in but somehow the removal of the entry was not successful. We are aware that if the statemap is not stored before the nifi restarts, it could lead to such circumstances, but in the recent occurrences of this issue, there were not nifi restart recorded or observed at the time when all those flowfiles started to queue in front of the processor group
 

 

Issue Links

• is duplicated by: NIFI-13281 Changing process group concurrency/outbound policy while data is within the group appears to result in stuck flow files
• links to: GitHub Pull Request #8951