Alicornutopia

I knew the upgrade caused memory problems but I didn’t realize they were so severe; that’s ridiculous.

I assume you know far more about it than me, but out of curiosity why’d you go to Ruby 2.1 instead of 2.3? It looks like they did even more work on memory since then, especially relevant to webapps, and it might make things better.

We just went one at a time; we're on 2.3.3 now. The problems started with 2.1 cause that's when the generational GC got added, but further upgrades haven't gotten me that far.

It's hard to tell cause I keep restarting the app to try things, but I've gotten it to stay around 380MB for at least a couple hours, so I reran the GC tuning and applied the latest configs. (Last time I ran it was pre some of my fixes, a Ruby version ago.) Going to see how it does overnight, fingers (and toes, and really any appendage I can come up with) crossed.

... Marri already answered, but, I was writing a post anyway so here it is. :P

We already jumped further, actually. We were doing it gradually because that's the recommended process for upgrading things like this – jump up a minor version, fix any problems that occur, jump up again, preferably with an extremely high test coverage.

We're yet to upgrade to Rails 4 - we upgraded a minor version and a couple of bugfix versions, but we're still on Rails 3. I'm not sure if Rails 4 improves memory usage – I haven't actually checked the patch notes – but it should improve security to a certain extent. Unfortunately it definitely does require changes to the code, and last I checked they were quite large changes, so we haven't done it yet. Upgrading Ruby did not, as far as I know, require any changes to code, fortunately.

*many hugs a marri*

[hugs]

We really don't pay you enough, Marri.

Is there anything users can particularly do/not do in order to avoid contributing to memory issues?

<3333

Hard to say, I can only sort of track down what's going on as it happens; there were two jumps last night (we're back to 445 :/) which appear to be first the Yandex bot crawling and second someone loading a couple large posts in flat mode. That second thing I can at least go poke at on the plane today.

that was probably me, i often do that to look up whether something has been mentioned before in a thread

Six hours at 350 so far, might be a record?

/stares at graphs with deeply excessive intensity

BEHAVE DAMNIT

does the search feature use less memory than loading a large post in flat mode? it seems like there are enough fields in the search feature to narrow it down to just a few threads, that might work for pedro?

Can we just have a "search this thread" feature?