Unique Jobs Not Running with Version 7.1.26

[marclennox]

I tried updating to 7.1.26 and now no unique jobs are actually running. Reverting to 7.1.25 resolves the issue.

[mhenrixon]

For me it is working fine:

worker | 2022-07-29T17:24:22.311Z pid=18000 tid=4s0 class=SlackNotifier jid=9ce8251c85717423e58aeb6d INFO: start
worker | 2022-07-29T17:24:22.326Z pid=18000 tid=4s0 class=SlackNotifier jid=9ce8251c85717423e58aeb6d elapsed=0.014 INFO: done

[mhenrixon]

Could you share some information about your setup @marclennox?

[marclennox]

I'll provide more detail when I get back to this, but my jobs defined as follows are not running in the newer version, but fine in the previous.

  sidekiq_options lock: :until_and_while_executing,
                  on_conflict: {
                    client: :log,
                    server: :reject
                  }

[marclennox]

They are all just going to the Dead queue

[mhenrixon]

Thanks for that, I'll try and replicate it in a local project!

[cgunther]

@marclennox any chance you're using redis-namespace?

I think the changes in v7.1.26 (namely #728) introduce a conflict with namespacing.

In v7.1.25, with no lock_timeout config set, when Locksmith#primed_async calls pop_queued, the wait was nil, so then pop_queued called rpoplpush.

In v7.1.26, pop_queued likely gets a timeout of 1 in the default scenario, thus triggering brpoplpush, which has a keyword argument, but I don't think redis-namespace properly handles it, which then leads to an error in the redis call, which seems to surface as the job conflicting with itself in this gem.

I didn't go terribly far down this rabbit hole, was only using namespacing in development, so ended up pulling it out completely, and that fixed things with v7.1.26. Jobs then ran fine in development and so far have run fine in production.

[mhenrixon]

Uff, redis-namespace strikes again... I can replicate the problem locally but fixing it I'm not even sure it is possible.

[mhenrixon]

This is the error message: Hash can't be coerced into Float

It seems like redis-namespace isn't compatible with newer redis versions. The redis-namespace uses the old version which took a float. Newer redis versions removed that argument:

def brpoplpush(source, destination, deprecated_timeout = 0, timeout: deprecated_timeout)

I don't want to use deprecated values if I can avoid it

[mhenrixon]

@cgunther @marclennox see if resque/redis-namespace#204 will be taken care of. There isn't that much I can do more than that.

[marclennox]

@marclennox any chance you're using redis-namespace?

I think the changes in v7.1.26 (namely #728) introduce a conflict with namespacing.

In v7.1.25, with no lock_timeout config set, when Locksmith#primed_async calls pop_queued, the wait was nil, so then pop_queued called rpoplpush.

In v7.1.26, pop_queued likely gets a timeout of 1 in the default scenario, thus triggering brpoplpush, which has a keyword argument, but I don't think redis-namespace properly handles it, which then leads to an error in the redis call, which seems to surface as the job conflicting with itself in this gem.

I didn't go terribly far down this rabbit hole, was only using namespacing in development, so ended up pulling it out completely, and that fixed things with v7.1.26. Jobs then ran fine in development and so far have run fine in production.

Indeed I'm using redis namespace. So I should abandon this to get around the problem?

[mhenrixon]

Mike Perham recommends strongly against redis namespace so I have come to work around it.

I'll see if I can fix it but given the type of error; I wonder how many other hidden errors are because of keyword arguments and redis namespace.