I think I may have narrowed it down a bit. I was unable to replicate the 502 errors on my mod/admin account, so I impersonated a user who had the problem and encountered it. When the user hits the submit reply button on a longer thread (4700 posts), the page sits on “saving” for a good 20 seconds before eventually failing to post with a 502 error.
Granting that user moderator status immediately fixes the problem. He was TL2, and granting TL3 did not fix it.
That is very interesting, cc @sam. Good sleuthing!
Since this is new, I am guessing this could be due to the consecutive reply check, will have a look later today
Given we have an easy bypass per:
Can you set max consecutive replies in your site settings to 0 and let me know what happens?
I think that MAY have fixed it-- it isn’t 100% reproducible, and I closed the thread I was testing in earlier, but re-opening it I was not able to reproduce. Will ask the users if they’re seeing any more 502s.
hmmm cause I just tested this query in data explorer on a 20k post topic and it is lightning fast.
SELECT user_id
FROM posts
WHERE deleted_at IS NULL
AND NOT hidden
AND topic_id = SOME_TOPIC_ID
ORDER BY post_number DESC
LIMIT 3
I would love to get to the bottom of this.
Happy to run whatever you want to help debug, but I’m not familiar with postgres or ruby so need some guidance.
Awesome, can you install Data Explorer Plugin and then run the query above substituting SOME_TOPIC_ID with the topic id of the problem topic?
All set. The topic I tested earlier:
3 results. Query completed in 0.5 ms.
and here is the output from the largest topic on the forum that’s not locked, 15k posts:
3 results. Query completed in 0.6 ms.
So seems really fast.
Also so far nobody has seen a 502 error, since we set that parameter to 0.
Thank you so much for 
ing with me.
I believe I just fixed the culprit here:
Any chance you can update to latest and re-enable the setting? Let me know if the issue is still gone?
Not sure how to update, dashboard says I’m already up to date and running git pull in /var/discourse says the same thing.
Just issue the rebuild from the command line and you should be good. You can time it this way too 
Did you try visiting /admin/upgrade?
OK, I updated (went to admin/upgrade manually, there was no link in the dashboard) and reverted the consecutive replies setting to 5. So far no errors, asked users to verify.
Thanks again for your help with this! Much appreciated!
There will only be a link from the dashboard when we release a new beta. We add code daily (hourly, even) - if we notified you for every single commit, your site would always say it’s out of date .
Makes sense, that’s why I tried a git pull earlier also. Anyway it’s upgraded now.
Thanks for your diligence in staying on top of this, we found a subtle but important issue as a result.