I recently upgraded to Agile 6.6.41 and I encountered this exact problem. It seems that your collation checks don't detect all special cases, in Czech collation, CH comes after H. I had to set the log to DEBUG, because there was no warning message like described above.

For PostgreSQL (I have 9.3), the workaround seems to be this (the workaround in the KB article only mentions changing the collation for the whole database which was not acceptable for me):

ALTER TABLE "AO_60DB71_LEXORANK"
   ALTER COLUMN "RANK" TYPE character varying(255) COLLATE pg_catalog."C";

OK. Wasn't even aware that you could change collation on a single table. This seems like a really dangerous solution/workaround to me.. Perhaps changing the collation of the entire database would/could be a better workaround but will effectively kill the ability to use local language in the application (which is kind of a bum).

The problem disappeared for us after applying the workaround (we have Swedish Finnish collation), but I'm not sure if it's a good practice to have a different collation on one specific table.

I think it would be better if Atlassian would fix the bug by rewriting Agile so that it can properly handle different collations,

I believe we are seeing the effect of this to. We run JIRA 6.3.11 and Agile 6.6.13 (with Swedish/Finnish collation). We experience (as an example) QuickCreateIssue execution times of some 20 seconds. Many other parts of the system performs slowly as well. We have a minima, out-of-the-box installation (quite fresh) on good, powerful hardware, with very few add-ons. This could not possibly be "normal" with JIRA, could it?

Is it really considered "resolved" if performance is affected? How much is performance affected and when?

This issue should now be resolved in JIRA Agile 6.6.13. What we have done to fix this is to put in place additional checks to figure out what collation is currently being used by the underlying database. If those checks tell us that the collation is non-standard, we apply additional sorting functions to our SQL queries when performing rank operations, to ensure that our ranking algorithm still performs as expected.

In this situation however, performance will be impacted, as applying these additional sorting functions slows down the operation of the database. There is also a message output in the logs when we detect this case:

****************************************************************************************************
 The database collation is set incorrectly for JIRA Agile. This can cause JIRA Agile to run slower. 
 If you are using PostgreSQL 9.0 or earlier, this may also cause ranking to fail.
 Please refer to the following documentation on the database configuration: https://confluence.atlassian.com/display/DOC/Database+Configuration
 If the problems persist, please refer to this KB on how to fix the database: https://confluence.atlassian.com/x/PAQWK
****************************************************************************************************

Regards,
JIRA Agile team

As stated in JRA-39970 it seems as though Jira will in the future require a specific collation. This is a poor solution to the problem in my opinion, as it excludes people who need a different collation.

Please confirm that this issue is not a problem with the default language setting for JIRA

Thanks for your comments on this issue so far. We will most likely tackle this issue as part of JRA-39970 and GHS-11162.

Regards,
JIRA Agile Team

Thanks Erik, we modified the bug summary to be more generic and also updated the diagnosis section.