Closed Bug 1286065 Opened 8 years ago Closed 8 years ago

Elasticsearch for bugzilla-dev

Categories

(bugzilla.mozilla.org :: Infrastructure, defect)

Product:
bugzilla.mozilla.org
Component:
Infrastructure
Version:
Production
Type:
defect
Priority:
Not set
Severity:
normal

Tracking

()

Status:
RESOLVED FIXED

People

(Reporter: dylan, Assigned: fubar)

Details

(Keywords: bugzilla-search) 

For furthering the work on the elasticsearch-based searches, I need a VM setup running elasticsearch 1.x and this needs to be accessible only from bugzilla-dev. This is just a precursor to BMO needing ES as well.
I need a VM with 6G of memory (ES's jvm uses only about 1G resident for the amount of data found on a dev db, but the total memory is just under 4G, so 6G seems safe). Two cores (if that's a thing I get to ask for),
and whatever a default amount of disk space is. The elasticsearch on disk DB is about 4G.

There should be net flows from the bugzilla-dev webheads and the admin node to this server, and it doesn't need to have any bugzilla code installed on it.

The version of elasticsearch should be 1.4.
Flags: needinfo?(klibby)
that's the same size that :ekyle's using for his stuff, so it's reasonable. we'll clearly want more nodes for prod, though. es 1.4 is ok; might have preferred more recent, but it's what we're using on DXR so I'm familiar with it.
Assignee: nobody → klibby
Flags: needinfo?(klibby)
aiming to have this in place by 10/18, per the tues meeting
es1.stage.bugs.scl3 is running, port 9200. cluster name is bugzilla_devstage. 

any *.bugs.scl3 host can get to 9200 already. if you need direct access before bug 1308941 is done you'll need to go through the admin node and then any other node first. 

let me know if you need any ES settings tweaked or have any other issues.
Status: NEW → RESOLVED
Closed: 8 years ago
Resolution: --- → FIXED
While working on prod, I added a zeus vip on the internal cluster (bugzilla-devstage-es.bugs.scl3) and /etc/hosts entries for es-dev and es-stage on the assumption that you're likely to do as you've done with the db hosts. 

The vip isn't any more useful than connecting directly atm, but if we need to add more nodes it'll help.
During patching, the cluster goes red and is sad. Even under normal circumstances it's yellow because ES is expecting to have replicas and there's no second node to put them on. We either need to tell ES to not have them or add more nodes. :dylan, any preference?
Flags: needinfo?(dylan)
None, I might have to change a configuration preference on my code because I think it specifies 2, but we don't need that for dev/staging.
Flags: needinfo?(dylan)
You need to log in before you can comment on or make changes to this bug.