Where communities thrive


  • Join over 1.5M+ people
  • Join over 100K+ communities
  • Free without limits
  • Create your own community
People
Activity
    Sergey Golitsynskiy
    @ic4f
    yep. Because the config will try to resolve as per schema; when path is absolute, os.path.join has no effect
    John Chilton
    @jmchilton
    We can fix that line by I'm worried other people will have paths relative to root in their configs?
    Or am I wrong?
    Sergey Golitsynskiy
    @ic4f
    they might
    we could change the schema? then tool_config would resolve wrt root. But won't the same issue potentially apply to any config file?
    we had this ambiguity all along: schema said 'will be resolved w.r.t. config_dir', but if you set it to a relative path, it was resolved w.r.t. root.
    John Chilton
    @jmchilton
    It would only resolve relative to the root - it wouldn't check both?
    Sergey Golitsynskiy
    @ic4f
    this is where it sets the intermediate path: https://github.com/galaxyproject/galaxy/blob/release_20.05/lib/galaxy/config/__init__.py#L292
    and this is where it resolves it wrt root: https://github.com/galaxyproject/galaxy/blob/release_20.05/lib/galaxy/config/__init__.py#L305
    it checked if value was not set.
    actually, i think we could check both - first config, then fall back to root.
    Marius van den Beek
    @mvdbeek
    :+1: though for not resolving relative to galaxy root by default. this default was ... slightly insane when all you wanted to do is have config/ live outside of Galaxy everything had to be absolute paths. And I don't think we want to have a fallback logic either, where admins accidentally load default config files if they made a mistake in the paths
    Sergey Golitsynskiy
    @ic4f
    So.. Resolve wrt config_dir, as per schema. Check if file exists. If yes - all good. If not: assume path was relative, and relative to galaxy root (because it would work before), so resolve wrt root, check if file exists; and if so - log warning? And if not - let galaxy raise the error?
    Marius van den Beek
    @mvdbeek
    I guess that's okay. We could maybe make that a depreciation warning?
    Sergey Golitsynskiy
    @ic4f
    we could. we do that for some stuff already.
    Marius van den Beek
    @mvdbeek
    :+1: if the deprecation warnings actually make it to the logs ... I think they might not
    Sergey Golitsynskiy
    @ic4f
    can do both
    Marius van den Beek
    @mvdbeek
    otherwise all caps!!! :laughing:
    Sergey Golitsynskiy
    @ic4f
    and a fair warning. Like next time you do it, you'll be eaten by a grue
    Sergey Golitsynskiy
    @ic4f
    Re: tool config file: here's a quick fix to dev, galaxyproject/galaxy#10200 . I'll have the real fix soon, but so that dev is not broken right now if used with that file..
    Sergey Golitsynskiy
    @ic4f
    Looks like a selenium test is failing consistently across multiple PRs (the failure is unrelated to the PRs, at least not to mine). galaxy_test.selenium.test_admin_app.AdminAppTestCase.test_admin_jobs_display
    John Chilton
    @jmchilton
    I'll take care of that - probably merged too quickly or merged something that had gotten out of date
    Thanks for the heads up
    Sergey Golitsynskiy
    @ic4f
    thanks!
    Pablo Moreno
    @pcm32

    Hi! We are unfortunately still getting 500 errors from the toolshed on installations:

    The galaxy instance logs show things like:

    tool_shed.galaxy_install.install_manager WARNING 2020-09-10 10:01:39,374 [p:51054,w:2,m:0] [uWSGIWorker2Core0] Error attempting to retrieve installation information from tool shed https://toolshed.g2.bx.psu.edu/ for revision 2704d80b7e91 of repository scanpy_find_cluster owned by ebi-gxa: HTTP Error 500: Internal Server Error

    and the latest ephemeris outputs for the same error:

    10:02:02 (419/607) Installing repository scanpy_find_cluster from ebi-gxa to section "hca_sc_scanpy_tools" at revision b456d129066d (TRT: 0:01:33.198307)
10:02:02     * Error installing a repository (after 0:00:01.816629 seconds)! Name: scanpy_find_cluster,owner: ebi-gxa, revision: b456d129066d, error: {"err_msg": "Uncaught exception in exposed API method:", "err_code": 0}

    We also see this other error in ephemeris:

    10:09:02     * Error installing a repository (after 0:00:00.566453 seconds)! Name: garnett_get_feature_genes,owner: ebi-gxa, revision: 3176cbad2800, error: {"err_msg": "Error attempting to retrieve installation information from tool shed https://toolshed.g2.bx.psu.edu/ for revision 3176cbad2800 of repository garnett_get_feature_genes owned by ebi-gxa: HTTP Error 500: Internal Server Error", "err_code": 500001}

    which is I think OOM related at the toolshed side?

    Marius van den Beek
    @mvdbeek
    yes
    selten
    @selten
    Is the toolshed using more RAM in Python3?
    Marius van den Beek
    @mvdbeek
    less, but we’ve overenthusiastically allocated less memory
    selten
    @selten
    Ah I see
    Marius van den Beek
    @mvdbeek
    it’s not exactly the TS either, most of the memory is hgwebdir, from where we’re serving merucrial repos
    Screenshot 2020-09-10 at 12.09.37.png
    selten
    @selten
    In my case it seems to use more RAM nowadays
    Thanks :), I see it did go down
    Marius van den Beek
    @mvdbeek
    Does anyone know if there is a good way to tell if a subprocess call has used posix_spawn or fork, short of money-patching subprocess ?
    Some of the Cannot allocate memory errors don’t really make sense if there’s still 2GB of memory free
    I know we’re not using posix_spawn because we’re passing cwd, but mercurial actually has a hidden commandline flag to indicate cwd
    (https://github.com/python/cpython/blob/3.8/Lib/subprocess.py#L1587)
    Nolan Woods
    @innovate-invent
    @mvdbeek does strace help at all?
    Are you working on the issue where hg doesn't appear in the uwsgi environment?
    For everyone else, I would appreciate some input on a new project I am undertaking. I am writing blend4go, or at least starting to. I would appreciate some input on the library interface.
    Nolan Woods
    @innovate-invent
    Should I try to emulate bioblend?
    Sergey Golitsynskiy
    @ic4f
    @eancelet (and anyone using a debugger on a galaxy instance). Back in June I suggested passing the --honour-stdin option to uwsgi by calling uwsgi directly (uwsgi --yaml config/galaxy.yml --honour-stdin) - that made debugging with pdb possible. There's a much better/simpler way to do that! Simply set honour-stdin: true in the uwsgi section in your galaxy.yml - that's it! Then you can use run.sh.
    4 replies
    John Chilton
    @jmchilton
    I'm going to attempt to turn get_quota into a sql query - I'm not sure it is a problem spot now but it probably would be if we generalized it to per-objectstore quota groups. Anyone ever attempt this or have any particular concerns? It seems at first glance like it should be all upsides in terms of performance but there might be some sort SA caching that is really beneficial and I'm just not seeing
    ^ @natefoo
    Dannon
    @dannon
    I think historically the query is just really expensive?
    (with sufficiently large user history)
    OH! get_quota, not usage.
    John Chilton
    @jmchilton
    get_quota doesn't depend on history sizes right? It is adding up all the quota adjustments across users and groups
    Yeah