ok i figured due to some reason yarn is trying to connect to marlon instead of worker-marlon. i think that might be the case if its tries public ip instead of internal ip.
interesting hdfs cli works as expected though reports the internal ips and worker-*
ok. the ip thing is fixed now but it now fails with user application exited with exit code 1.
alastairp
CatQuest: thanks! blank spaces is OK, the only problem we might have is an empty string. Anything that is actually represented by characters is fine 👍
sumedh joined the channel
reosarevok
bitmap, yvanzo: for MBS-1658, I was thinking at least one of the places to add a comment to the entry should be from the list itself
bitmap, yvanzo: So that last column should have like an edit icon somehow, and I guess allow inline editing that would get sent to the DB? Do you know if we do anything like that anywhere else?
Or if you think that's a bad idea, how would you do it?
alastairp
CatQuest: hah, nice
that's not a problem either though, but I can see how it could be a problem
Mr_Monkey: alastairp : so after more experimenting last night, I'm able to get rid of the min/max ts cont aggs by simply creating a 5 days cont agg with compound index on user/listened_at. Thats 18M rows less for starters.
and I think we can replace those with month and year cont aggs for the graphs you two would like.
alastairp
right. get all data from the same table?
ruaok
yeah, it was already there. just the index was missing to make it faster.
alastairp
sweet, if a month and year aggregate is possible then that sounds like it should be perfect
great
ruaok
basically we swapped doing a table scan on the DB with an index scan. not sure we can do much better than that -- but with increased cache times, this should work well.
alastairp
_lucifer: ^ remember how I told you to add indexes to tables where you want to select some data?
_lucifer
yup, i'll keep it in mind :D
ruaok
its the rookie mistake that keeps on giving. #going25yearsstrong
so, lets do all the loading of data (mapping, incrementals) then we can fire off some jobs.
that makes sense.
very very good.
_lucifer
two things left to do. one is define memory defaults and second is update the new configuration in syswiki.
monitoring this cluster is easier than the docker one. one tunnel is sufficient
ruaok
that was exactly the goal.
and each server is being monitored by all of zas' magic.
_lucifer
:D
alastairp: available to talk about the GH actions PR?
ruaok, zas: do you know if any service we run on j5 might listen on port 5666?
zas
nagios
well, its agent
_lucifer
👍 thanks!
sumedh has quit
sumedh joined the channel
zas, i saw a commit in syswiki renaming germaine to jermaine so wanted to let you know that i noticed /etc/hosts on jermaine still contains has a couple of entries referring germaine.
bitmap, yvanzo ^ would really appreciate some feedback on whether the way I'm approaching this seems sensible, improvements etc, before I keep working on other lists
Hello everyone. I would like to ask about Lucene Search syntax of the musicbrainz database. What kind of instance is it running on? If i would like to have a musicbrainz database (mbdata) with ElasticSearch instance, how do you recommend to integrate these two. I just started looking into ElasticSearch but what I found out i would need some kind of
data set to import it to ElasticSearch (*.json for example). Do you have some kind of method to import musicbrainz database into a Lucene Search intance, a data set I can use, or i should generate it myself? How do you keep it updated?
ruaok
hi scory!
why must is be elasticsearch?
because we have a perfectly working search infrastructure that you can use without having to reinvent the wheel.
scory
That infrastructure currently doesn't support what I need, last time I was here that was the conclusion for me. That's the reason I am currently running a mbdata server locally, and can run graphql queries against it, with batching. But I would like to implement an ElasticSearch instance on graphql. But currently i am just investigating.
reosarevok: I can't think of anything else like that offhand, but doesn't seem like a bad idea. we could add a small endpoint to /ws/js for it
sumedh joined the channel
vardan has quit
adhi001 joined the channel
adhi001
Sorry ruaok , I was sick the last week and was not able to submit a proposal for GSoC. Still part of the community :)
ruaok
oh, bummer. that sucks. at least you're better, right?
adhi001
yeah
Thank you for your concern
alastairp
_lucifer: hi, sorry - had a hectic day. still around?
_lucifer
alastairp: hi! no worries. yup, i am available.
alastairp
so I was suggesting using test.sh in the actions?
_lucifer
yes
alastairp
so we already have things like `./test.sh -b` to build, and `test.sh -u` to bring up containers
test.sh fe to run frontend tests
_lucifer
yes there's also test.sh spark
alastairp
great, so it sounds like it's probably a good fit that we can use the actions files for specifying the orders in which to run things, but reusing test.sh for the actual commands allows us to share 100% test code between local developement and CI, right?
_lucifer
should we use separate build steps? like there's a ./test.sh -u to just bring up supporting containers. or should we just to do ./test.sh which does it in one go
alastairp
but we need to separate pull / cache / build / run, in CI, right?
_lucifer
yes, mostly. we'll still have to pull manually
that step won't change
alastairp
one question - if a test generates files (e.g. the junit xml), will it be cached? Or does the cache action only cache docker layers?
_lucifer
i'll need to check that.
i expect docker layers only but we can confirm it by generating some files and looking at the actions output
alastairp
it seems that satackey/action-docker-layer-caching works explicitly on layers (looking at the output, it uses docker commands to generate the archives)
so yeah, ./test.sh pull; restore cache; run test.sh; save cache
the junit action works fine but as i mentioned it might comment excessively
alastairp
neat. did you see what happens if one fails? does it only update the comment or does it also add an annotation to the failing test?
and it'll make a new comment on every push (i.e. even if they all pass?)
_lucifer
no i haven't, let me do that right now.
yes
alastairp
or only if the results of the test run change?
interesting
you're right that this could get a bit annoying
_lucifer
it'll hide the existing one and add a new one
for LB there are going to be 4 comments on each push
alastairp
oh, that's quite annoying. merging tests together would help (e.g. get it down to 2), but I suspect that this might be too much
one thing that ruaok was suggesting back on jenkins was that it seems stupid to run tests on _every_ push, perhaps there could be a way to run them less often. once a day? on request based on a comment? just before merge?
ruaok
anything, really.
alastairp
let's not spend too much more time on this, but perhaps there is an action or a flag for `on:` that lets us decide to run them less often