@voidswitch opened this Issue on October 13th 2014


currently we encountered a strange issue with pages listed in actions->pages, but not if using a segment (which should include this page). Investigating the issue I found duplicate entries in piwik_log_action.

idaction name hash type url_prefix
4114714 example.com/somepage 652463546 1 1
4114715 example.com/somepage 652463546 1 1

In piwik_log_link_visit_action only idaction 4114715 is referenced, and displayed with views in actions->pages; creating a segment matching example.com/somepage gets 0 views for this page.

Something to consider is that, we use 2 balanced Piwik installations for tracking, a setup which should be working fine, but also has the potential to create this kind of issue as type, hash is not unique.

@voidswitch commented on October 13th 2014

Looking through ./core/Tracker/TableLogAction.php there seems to be the possibility, that two concurrent request with the same URL could result in duplicate entries.

Consider loadIdsAction executed in parallel on two installations with the same input. There should be a small time window, where both read the current state, create their list of items to be inserted, then insert the same item with different idactions, so that valid entries are created.

@mattab commented on October 17th 2014 Member

Thanks for the report, we should really fix this issue (although I'm not sure how to fix it).

@voidswitch commented on October 23rd 2014

If I understand correctly it doesn't make sense to have multiple rows in _logaction with the same name/type combination, so adding a unique constraint to _logaction for name/type or hash/type and changing insertNewIdsAction to use insert ignore or insert ... on duplicate should fix this.

Although I'm not sure about cleanup of existing data. If duplicates need to be deleted we have to cleanup _log_link_visitaction with about 700,000,000 entries

@voidswitch commented on October 27th 2014

Another possibility to workaround the issue of multiple Piwik instances writing data seems to be something like

INSERT INTO log_action ( name, hash, type, url_prefix )
SELECT * FROM (SELECT ?, CRC32(?), ?, ?) AS tmp
    SELECT name FROM log_action WHERE name = ? and type = ?
) LIMIT 1;

This statement should avoid inserting an entry with the same name and hash as an existing entry without using an UNIQUE constraint.

Please note: I was not able to test this statement (syntax or if it would fix the problem) due to time constraints ;)

@mattab commented on October 28th 2014 Member

@voidswitch thanks for your comments that's useful. Hopefully we get to this ticket in coming weeks as it's causing other bugs such as #4595

@integrationtester commented on October 30th 2014

for a cleanup script, you can't just delete the 'duplicate' entries, you have to merge them:

select idaction,name,count(*) from piwik_log_action as action, piwik_log_link_visit_action as linker where linker.idaction_url_ref=action.idaction group by action.idaction order by name;

for each of the names which have multiple idactions:
(in my case i had upwards of 10 ids for the same url)

say i have multiple idactions: 63,90,134,213,180

update piwik_log_link_visit_action set idaction_url_ref=63 where idaction_url_ref in (90,134,213,180);

update piwik_log_link_visit_action set idaction_url=63 where idaction_url in (90,134,213,180);
delete from piwik_log_action where idaction in (90,134,213,180);

@demaya commented on December 10th 2014

Did you solve the problem already? For my most visited page I have no transitions since 2013 :-1:

I select like 3500 rows with the above mentioned SQL-select statement! It's not that I can manually clean them up...

@fpoepleu commented on January 12th 2015

It would be really great if this were solved. We have the same problem and our databases date back to 2011.
Are there any news?

@voidswitch commented on February 9th 2015

Just want to say thanks for the great work!

@mattab commented on February 9th 2015 Member

Thanks @voidswitch I also want to say great work to @diosmosis and the team!

This Issue was closed on February 9th 2015
Powered by GitHub Issue Mirror