Solr tweaks #1466

jywarren · 2017-06-15T20:50:17Z

all tests pass -- rake test:all
code is in uniquely-named feature branch, and has been rebased on top of latest master (especially if you've been asked to make additional changes)
pull request is descriptively named with #number reference back to original issue

Please be sure you've reviewed our contribution guidelines at https://publiclab.org/wiki/contributing-to-public-lab-software

We have a loose schedule of reviewing and pulling in changes every Tuesday and Friday, and publishing changes on Fridays. Please alert developers on plots-dev@googlegroups.com when your request is ready or if you need assistance.

Thanks!

jywarren · 2017-06-15T20:51:18Z

Again back to https://github.com/sunspot/sunspot#manually-adjusting-solr-parameters

Based on console testing, it could also work to remove &defType=edismax, but I'm not sure how to do that as easily in rails compared to just assigning a blank string to params[:qf].

PublicLabBot · 2017-06-15T20:59:55Z

	1 Warning
⚠️	It looks like you merged from master in this pull request. Please rebase to get rid of the merge commits – you may want to rewind the master branch and rebase instead of merging in from master, which can cause problems when accepting new code!

	3 Messages
📖	@jywarren Thank you for your pull request! I’m here to help with some tips and recommendations. Please take a look at the list provided and help us review and accept your contribution!
📖	This pull request doesn’t link to a issue number. Please refer to the issue it fixes (if any) in the body of your PR, in the format: `Fixes #123`.
📖	It looks like you haven’t marked all the checkboxes. Help us review and accept your suggested changes by going through the steps one by one. If it is still a ‘Work in progresss’, please include ‘[WIP]’ in the title.

Generated by 🚫 Danger

jywarren · 2017-06-15T21:03:40Z

OK, here i'm stuck -- bc based on my testing on the Solr console on staging, we need to set params[:qf] = "" -- but that seems to break the tests using embedded Solr.

I can try some other variations, like params[:qf] = "''" -- empty quotes -- but maybe we should instead try removing the edismax part...

jywarren · 2017-06-15T21:15:32Z

Trying .delete(:defType)... odd but who knows

jywarren · 2017-06-15T22:05:49Z

Should be able to test this with http://staging.laboratoriopublico.org/searches/test/test?q=stk500_recv

jywarren · 2017-06-15T22:12:49Z

Unfortunate... @icarito -- any inspirations?

In summary:

We know the exact Solr query we want, we just can't get Sunspot to create it.

Alternatively, we aren't sure why the default Sunspot-generated query (with qf=title_text+body_text+comment_text and defType=edismax) doesn't work -- removing either of those portions of the query gets us full results.

Actually, to be specific, using adjust_solr_params to change qf to nil works properly on the staging server, but it causes the embedded Solr to return no results.

jywarren · 2017-06-18T16:22:03Z

Added more detail to the sunspot issue and opened one on Stack Overflow: https://stackoverflow.com/questions/44617017/difficulty-getting-more-than-exact-field-match-results-on-sunspot-solr-with-qf

jywarren · 2017-06-18T17:16:31Z

Trying to set params[:defType] to an empty string, which on rails console testing resulted in:

irb(main):011:0> s = Node.search { fulltext('balloon');adjust_solr_params { |p| p[:defType] = '' } }.results
  SOLR Request (109.0ms)  [ path=select parameters={fq: ["type:Node"], q: "balloon", fl: "* score", qf: "title_text body_text comments_text", defType: "", start: 0, r
ows: 30} ]

And full results...

jywarren · 2017-06-18T17:36:54Z

OK, another variation which worked on external Solr but did not pass embedded solr tests:

Failure:
  <[]> expected to be != to
  <[]>.
test_should_get_search_test_action(SearchesControllerTest)
test/solr/searches_controller_test.rb:23:in `block in <class:SearchesControllerTest>'

Sebastian -- the alternative here is that we could simply drop support of the embedded gem sunspot_solr engine (which is for development only anyways) and spin up the containerized Solr in Travis.

What would that take? Changes to:

uncomment the solr container in docker-compose.yml (for purposes of Travis only)
repoint /config/sunspot.yml to the container (just the port number, i guess)
would the solr-specific test.rake lines require tweaks so that the commands to start/stop the solr engine would go to the containerized solr, not the embedded solr?

This seems like a way forward even if we don't figure out what's going on. I suspect the embedded solr is just an older version and is not responding to the same queries in the same ways. Maybe it's not using edismax queries, for example.

jywarren · 2017-06-18T17:46:26Z

Actually i like this plan a lot. It's closer to how we'll be deploying anyways.

jywarren · 2017-06-19T01:10:28Z

Attempting these steps now -- not 100% sure of the sunspot.yml or .travis.yml or test.rake changes, but let's see how it does!

jywarren · 2017-06-19T02:26:11Z

Seems close -- solr container starts up, but it seems to run /after/ bower install for some reason, though i tried adding 40 seconds wait. The last thing that runs is WeBrick starting Rails, but i never see rake db:migrate run, or tests.

jywarren · 2017-06-22T18:58:05Z

Hmm, so what is going on here? Can you push up a copy without logging on, not a force push? What's the specific error?

icarito · 2017-06-22T21:24:13Z

Logging is the only place where we are seeing the actual error.
rake db:migrate is run without errors: https://travis-ci.org/publiclab/plots2/builds/245957757#L2999 check the raw log...

icarito · 2017-07-07T16:00:02Z

Ok, so the reindex appears to be working fine. I'll try to reproduce in a local environment.

jywarren · 2017-07-11T17:03:05Z

Hi, Sebastian -- how's this going?

icarito · 2017-07-11T17:30:03Z

I don't why, I thought we had solved it. Likely needs a last push, will review it now. Sorry!

icarito · 2017-07-11T19:18:15Z

I've merged this into unstable branch so that we can use jenkins to track this.
This made me notice that in our current setup, staging and production are sharing the same pad.publiclab.org Solr index! At one point I stopped the Solr container in pad.publiclab.org and experienced a 500 error in publiclab.org/dashboard! So I think we need to doublecheck our fallbacks.

jywarren · 2017-07-11T19:27:47Z

Yikes! So staging should have its own index, right? Is that then in sunspot.yml?

…

On Tue, Jul 11, 2017 at 3:18 PM, Sebastian Silva ***@***.***> wrote: I've merged this into unstable branch so that we can use jenkins to track this. This made me notice that in our current setup, staging and production are sharing the same pad.publiclab.org Solr index! At one point I stopped the Solr container in pad.publiclab.org and experienced a 500 error in publiclab.org/dashboard! So I think we need to doublecheck our fallbacks. — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub <#1466 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AABfJ9AIBdr0KrCbRyTJVI33e4CjvN9kks5sM8p3gaJpZM4N7ue8> .

icarito · 2017-07-11T19:30:42Z

Yes! I've closed the firewall at pad.publiclab.org to only accept connections from publiclab.org (dropping requests from tycho.media.mit.edu). That should ensure isolation from staging. I'm going to set SOLR_URL in Jenkins, it should override the config in sunspot.yml.

icarito · 2017-07-11T19:39:21Z

In fact, having firewalled tycho from pad.publiclab.org should affect staging and allow to debug the fallback. Checking it...

icarito · 2017-07-11T19:54:39Z

I'm not seeing the staging server fail after restricting access to pad.publiclab.org. I'll check it further tonight.

icarito · 2017-07-11T22:18:13Z

My fault, I had put the wrong address in SOLR_URL env variable in jenkins config. It appears to be working fine after changing it.

icarito · 2017-07-11T22:28:11Z

I think staging is working except without access to Solr. I'll have a closer look tonight.

icarito · 2017-07-12T02:03:08Z

Okay I was trying to reindex tonight and got this:

icarito@tycho:/srv/plots_testing/plots$ RAILS_ENV=production SOLR_URL=http://solr:8983/solr/ docker-compose exec web rake DISABLE_SOLR_CHECK=1 sunspot:solr:rei
ndex
rake aborted!
Errno::ETIMEDOUT: Connection timed out - connect(2) for "pad.publiclab.org" port 8983
(eval):2:in `post'
Tasks: TOP => sunspot:solr:reindex => sunspot:reindex
(See full trace by running task with --trace)
icarito@tycho:/srv/plots_testing/plots$

This is because the firewall is set to DROP and I guess the Solr-check is timing out instead of immediately failing, taking a long time. I'll make a patch to enable Solr container in docker-compose.yml and also try setting the firewall to REJECT, to see if the Solr-check will work better in that case.

…ots2 into solr-no-query-fields

icarito · 2017-07-13T01:48:27Z

@jywarren sincerely I've spent a lot of time on this one and I can't tell why it's failing. I would appreciate your help.

jywarren · 2017-07-13T02:17:31Z

Ok we can look at this tomorrow! Thanks for spending time on this.

My first thought is can we view and reproduce this query directly on solr console on staging.

jywarren · 2017-07-13T18:30:39Z

http://staging.laboratoriopublico.org/searches/test?q=spectrometer works

icarito · 2017-07-13T19:22:10Z

Today Jeff and I worked on this, configuring the staging environment to run from the test RAILS_ENV, and discovered that the indexer was indexing the seed data instead of the test data. We're attempting to reindex after the original tests have been run, to see if it will include the test data.

icarito · 2017-07-13T19:36:58Z

YESS!

jywarren · 2017-07-13T19:43:02Z

HOORAAYYYYYYYY

lets get this on staging fast!

jywarren · 2017-07-13T19:43:25Z

Ready to merge, right? if so, can you hit merge and i'll push to staging?

icarito · 2017-07-13T19:44:29Z

Yes, please!

…

On 13/07/17 14:43, Jeffrey Warren wrote: Ready to merge, right? if so, can you hit merge and i'll push to staging? — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub <#1466 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AAMMSxZlxbVkiuOYcMW7bXRsIo5-4W1Uks5sNnNdgaJpZM4N7ue8>.

icarito · 2017-07-13T19:53:18Z

Note that staging will be broken until I restore the staging database from a dump. I can do it later today.

icarito · 2017-07-14T06:37:59Z

Staging is back with a reindexed solr container, all lights green.

jywarren mentioned this pull request Jun 15, 2017

Planning issue for Search integration #1463

Closed

17 tasks

icarito mentioned this pull request Jun 16, 2017

Different behaviour of embedded Solr vs Solr container? sunspot/sunspot#863

Closed

jywarren added 3 commits June 18, 2017 21:14

.getDocs

2691fd4

changed to .items

1c5961f

attempt to run solr in container in travis

b02f033

jywarren force-pushed the solr-no-query-fields branch from 86f9ac1 to b02f033 Compare June 19, 2017 01:16

jywarren added 3 commits June 18, 2017 21:25

docker-compose command tweak

7c25a2e

another tweak attempt

c5a63bb

longer sleep/

765ffad

icarito force-pushed the solr-no-query-fields branch 5 times, most recently from 301b4a1 to 418233c Compare June 22, 2017 18:14

Fix Solr container startup.

f6ede88

icarito force-pushed the solr-no-query-fields branch from 418233c to f6ede88 Compare June 22, 2017 20:49

icarito mentioned this pull request Jul 12, 2017

Reenable Solr and include Solr sunspot schema #1523

Merged

3 tasks

icarito and others added 3 commits July 12, 2017 10:28

Merge branch 'master' into solr-no-query-fields

ebde068

Tweaks. Query fields solved, trying to make sure tests pass.

dc3c853

Merge branch 'solr-no-query-fields' of http://github.com/publiclab/pl…

266c256

…ots2 into solr-no-query-fields

jywarren added 3 commits July 13, 2017 14:37

Update .travis.yml

b177798

Update test.rake

5e2aaee

Update .travis.yml

94db2bc

jywarren merged commit 840bc36 into master Jul 13, 2017

emilyashley deleted the solr-no-query-fields branch January 15, 2020 21:35

Solr tweaks #1466

Solr tweaks #1466

Conversation

jywarren commented Jun 15, 2017

jywarren commented Jun 15, 2017 • edited Loading

PublicLabBot commented Jun 15, 2017 • edited Loading

jywarren commented Jun 15, 2017

jywarren commented Jun 15, 2017

jywarren commented Jun 15, 2017

jywarren commented Jun 15, 2017 • edited Loading

jywarren commented Jun 18, 2017

jywarren commented Jun 18, 2017

jywarren commented Jun 18, 2017

jywarren commented Jun 18, 2017

jywarren commented Jun 19, 2017

jywarren commented Jun 19, 2017

jywarren commented Jun 22, 2017

icarito commented Jun 22, 2017

icarito commented Jul 7, 2017

jywarren commented Jul 11, 2017

icarito commented Jul 11, 2017

icarito commented Jul 11, 2017

jywarren commented Jul 11, 2017 via email

icarito commented Jul 11, 2017

icarito commented Jul 11, 2017

icarito commented Jul 11, 2017

icarito commented Jul 11, 2017

icarito commented Jul 11, 2017

icarito commented Jul 12, 2017

icarito commented Jul 13, 2017

jywarren commented Jul 13, 2017

jywarren commented Jul 13, 2017

icarito commented Jul 13, 2017

icarito commented Jul 13, 2017

jywarren commented Jul 13, 2017

jywarren commented Jul 13, 2017

icarito commented Jul 13, 2017 via email

icarito commented Jul 13, 2017

icarito commented Jul 14, 2017

jywarren commented Jun 15, 2017 •

edited

Loading

PublicLabBot commented Jun 15, 2017 •

edited

Loading

jywarren commented Jun 15, 2017 •

edited

Loading