https://github.com/ceph/shaman

There are three VMs in the OVH CI region that make up shaman.

• shaman.ceph.com is just a load balancing VM. Accesses are proxied to either 1.shaman.ceph.com or 2.shaman.ceph.com with an 'upstream shaman' clause in /etc/nginx/nginx.conf, which is then referred to by the site config for shaman.ceph.com.
• 1.shaman.ceph.com is the primary shaman node that has the postgres DB with all the repo information
• 2.shaman.ceph.com is a READ ONLY backup in the event 1.shaman.ceph.com goes down
• 2.shaman.ceph.com can handle write requests because pecan, the web framework, is also aware of the primary/hot standby configuration, and so will redirect writes to 1.shaman.ceph.com on its own, if they appear.

User: ubuntu (or infra admin username)
Key: CI private key (or your private key)
Port: 2222

systemctl start|stop|restart|status shaman

1. If needed, copy deploy/playbooks/examples/deploy_production.yml to deploy/playbooks/
   1. Get and set the following credentials. These can be found in /opt/shaman/src/shaman/prod.py on 1.shaman.ceph.com
      1. api_user
      2. api_key
      3. rabbit_host
      4. rabbit_user
      5. rabbit_pw
      6. github_secret
2. Run the playbook (see below)

**It is extremely important that the postgres tag is skipped**

Set --limit to one node at a time to avoid disrupting the CI or lab testing.

ansible-playbook --tags="deploy_app" --skip-tags="postgres,nginx" --extra-vars="master_ip=158.69.71.144 standby_ip=158.69.71.192" deploy_production.yml --limit="1.shaman.ceph.com,2.shaman.ceph.com"

I needed to determine what percentage of jobs were running on static vs. ephemeral slaves. Alfredo wrote a python script to pull this data out of the shaman database. This script totals how many jobs ran on static vs. ephemeral slaves over a 2 week period (since that's how long we keep dev builds).

Doing this on 2.shaman.ceph.com ensures you're in a read-only capacity.

1. ssh 2.shaman.ceph.com
2. cd /opt/shaman/src/shaman
3. Copy the script below to two_week_stats.py

   import datetime
   
   from shaman import models
   from shaman.models import Build, Project
   models.start_read_only()
   
   
   
   def report():
       two_weeks = datetime.datetime.utcnow() - datetime.timedelta(days=15)
       
       ceph_project = models.Project.filter_by(name='ceph').one()
       builds = Build.filter_by(project=ceph_project).filter(Build.completed > two_weeks).all()
   
       ovh_builds = {}
       irvingi_builds = {}
       braggi_builds = {}
       adami_builds = {}    
       rest_of_the_world = {}
       ovh_count = 0
       irvingi_count = 0
       braggi_count = 0
       adami_count = 0
   
       for build in builds:
           node_name = build.extra['node_name']
           if '__' in node_name:
               mapping = ovh_builds
               counter = ovh_count
           elif 'slave-' in node_name:
               mapping = irvingi_builds
               counter = irvingi_count
           elif 'braggi' in node_name:
               mapping = braggi_builds
               counter = braggi_count
           elif 'adami' in node_name:
               mapping = adami_builds
               counter = adami_count
           else:
               mapping = rest_of_the_world
       
           try:
               mapping[node_name] += 1
           except KeyError:
               mapping[node_name] = 1
       
       for mapping in [ovh_builds, irvingi_builds, braggi_builds, adami_builds]:
           count = 0
           for key, value in mapping.items():
               print key, value
               count += value
   	print "TOTAL: %s" % count
           print "="*60
           print

4. /opt/shaman/bin/pecan shell --shell ipython prod.py
5. Then

   In [1]: from shaman import models
   
   In [2]: models.start_read_only()
   
   In [3]: import two_week_stats
   
   In [4]: two_week_stats.report()

HYPOTHETICALLY if a repo/build got pushed to shaman that contains an embargoed security fix, you can delete the entries from shaman's DB. The packages will still be on chacra servers but shaman won't know about them. You can always delete them from chacra too if necessary.

ssh 1.shaman.ceph.com
sudo su - postgres

postgres@1:~$ psql -d shaman
psql (9.5.23)
Type "help" for help.

shaman=# \dt
             List of relations
 Schema |      Name       | Type  | Owner  
--------+-----------------+-------+--------
 public | alembic_version | table | shaman
 public | archs           | table | shaman
 public | builds          | table | shaman
 public | nodes           | table | shaman
 public | projects        | table | shaman
 public | repos           | table | shaman
(6 rows)

shaman=# delete from public.builds where sha1 = 'f73b19678311b996984c30e7c0eb96a22ffa29ce';
DELETE 6

shaman=# select id from public.repos where sha1 = 'f73b19678311b996984c30e7c0eb96a22ffa29ce';
   id   
--------
 197001
 197010
 197011
 197012
 197030
 196999

shaman=# delete from public.archs where repo_id = '197001';
DELETE 1
shaman=# delete from public.archs where repo_id = '197010';
DELETE 2
shaman=# delete from public.archs where repo_id = '197011';
DELETE 2
shaman=# delete from public.archs where repo_id = '197012';
DELETE 2
shaman=# delete from public.archs where repo_id = '197030';
DELETE 2
shaman=# delete from public.archs where repo_id = '196999';
DELETE 1
shaman=# delete from public.repos where sha1 = 'f73b19678311b996984c30e7c0eb96a22ffa29ce';
DELETE 6

There's a custom Nagios check in place that queries the /api/nodes/next endpoint.

This check is in place to make sure the postgres database is writeable. An incident occurred in 2019 where OVH rebooted all 3 shaman-related VMs at the same time and the DB was read-only for an unknown reason.

root@nagios:~# cat /usr/lib/nagios/plugins/check_shaman 
#!/bin/bash
# Checks shaman /api/nodes/next endpoint

if curl -s -I -u XXXXX:XXXXX https://${1}/api/nodes/next | grep -q "200 OK"; then
    echo "OK - Shaman /api/nodes/next endpoint healthy"
    exit 0
else
    echo "CRITICAL - Shaman /api/nodes/next endpoint failed"
    exit 2
fi