Hi Bence, Joe

Thank you guys for your fast response.

@Bence you are right, the whole issue seems to be related to nested flows which 
we try to sync. We are relying heavily on this as we are doing the integration 
on one NiFi system and then sync it to production via the NiFi Registry. Our 
parent flow contains multiple nested flows.

Ok, as NIFI-10874 will be included in the next release we will skip the 1.19.0 
and wait for 1.19.1 / 1.20.0. Sadly it will probably take more than a few days 
until we see the next release... Do you have an idea why the flow with the 
nested flow could lose the storageLocation? Because, we just change a small 
thing in one of the nested flows and suddenly one of the other nested flows 
have been losing it’s storageLocation in the parent flow and from this point 
the whole flow was broken. Now we are scared to change anything as it could 
brake anytime again.

Cheers Josef



From: Simon Bence <[email protected]>
Reply to: "[email protected]" <[email protected]>
Date: Tuesday, 29 November 2022 at 13:31
To: "[email protected]" <[email protected]>
Subject: Re: NiFi Registry Bug which brakes the flow sync with NiFi 1.18.0 (and 
same version of the registry) on nested flows

Hi Josef,

Thanks for your patience!

I took a deeper look on what you were writing. In general, this is a sign for 
the case where the registry client cannot find the nested flow. As far as I 
understood in your case this happens when you are having a sync. Based on this 
you may be hitting https://issues.apache.org/jira/browse/NIFI-10874 which is 
not part of NiFi 1.19.0. The next release should contain the fix that addresses 
your issue.

Regards,
Bence


On 2022. Nov 28., at 17:48, Simon Bence 
<[email protected]<mailto:[email protected]>> wrote:

Hi Josef,

Thank you for raising the attention to this.

The storageLocation is a new concept to generalise the information used to 
locate nested flows in a versioned flow. In general it can be there and does 
not cause issues, as of now NiFi falls back the previous way to locate nested 
flows. As Joe mentioned, there is a fix in 1.19, but I need to double check if 
it relates to this situation. Please give me some time, I will find you back.

Regards,
Bence


On 2022. Nov 28., at 17:17, Joe Witt 
<[email protected]<mailto:[email protected]>> wrote:

Josef

Sorry for the challenges you've hit there.  I do think in 1.18 we had a 
bug/regression as we refactored our registry client.  That issue should be 
resolved in 1.19 which just went live thanks to 
https://issues.apache.org/jira/browse/NIFI-10787.  However, I am not positive 
if this will solve the scenario you've hit now but please if able try it out.

Thanks

On Mon, Nov 28, 2022 at 8:50 AM <[email protected]> wrote:
Hi guys



We had the following bug twice already and it broke the whole sync with the 
NiFi registry for the Flow/PG. First time was directly after we have upgraded 
from NiFi 1.15.3 to 1.18.0, but we ignored it as we thought it could be because 
of the upgrade, however it occurred again after a few NiFi Registry commits on 
NiFi 1.18.0… The error was the following when we tried to change the flow 
version or when we tried to start from scratch with that version from the NiFi 
Registry, so the version was broken in the NiFi Registry:



<image001.png>





We investigated the last NiFi Registry Commit and we saw in our GIT repo (we 
sync the “flow_storage” to GIT) on line 4847 that NiFi removed a 
“storageLocation” from another PG which makes no sense at all. We changed 
nothing there and especially why should NiFi remove only the storageLocation 
line… We have one specialty, as we have nested NiFi Registry flows, so one of 
the flows where the storageLocation has been removed was such a nested flow.





<image002.png>





Luckily we were able to resolve the error. We tried to add the line and commit 
it to GIT plus we dropped the the database to repopulate the DB, however it was 
broken again after a commit from NiFi. So we tried to manually create a new 
fake version on bucket.yml in the corresponding bucket folder and added as well 
the line in the snapshot again. We dropped then the DB and restarted NiFi 
Registry and voilà it was working again.



However it was a nightmare to get it working again as the flow was completely 
broken, we couldn’t checkout the affected version at all.



Any thoughts on this? Shall I fill a Jira Ticket? The problem is, we can’t 
really reproduce it, it looks like it happens randomly. As you could imagine, 
we can’t share our template as it contains a lot of confidential material.



Cheers Josef




Attachment: smime.p7s
Description: S/MIME Cryptographic Signature

Reply via email to