This task is to pick a database for Nodepool
For continuous integration purposes, we are setting up a python based daemon named Nodepool. It maintains a pool of instances on the labs infrastructure and refill the pool as instances are consumed by Jenkins jobs.
To track status of instances and images in labs, Nodepool relies on a database backend. During the proof of concept build, I manually installed a MySQL database on the machine (labnodepool1001.eqiad.wmnet). When reinstalling fully from puppet, I found out I forgot about the database part and the service is obviously no more working.
I do not expect much queries per seconds to happen though I have no metric to back up my claim. It is definitely less consuming than Phabricator or Gerrit might be.
Nodepool hold a database connection for its internal process as well for each instances spawned in labs. Seems we might want up to 100 connections for a start. I am not sure how much of an issue it can be for our databases.
Once we have a database and credentials, the Nodepool services takes care of populating the database schema and performing schema changes. The suggested installation process is:
CREATE USER 'nodepool'@'localhost' IDENTIFIED BY '<password>'; CREATE DATABASE nodepooldb; GRANT ALL ON nodepooldb.* TO 'nodepool'@'localhost'; FLUSH PRIVILEGES;
Nodepool uses a pure python MySQL client library (PyMysql). The credentials are configured in Nodepool file in operations/puppet.git:
modules/nodepool/templates/nodepool.yaml.erb: dburi: 'mysql+pymysql://nodepool:nodepool@localhost/nodepool'
Nodepool administrators can thus reuse those credentials if they need to manually interact with the db though anything we need is provided by the Nodepool command line client.
Potentially we can either:
- colocate the database with the nodepool service (on labnodepool1001.eqiad.wmnet)
- host the DB on one of our misc db hosts
The server is labnodepool1001.eqiad.wmnet, might want to have the DB as close to it as possible if that matters.
Unrelated. It is a standalone daemon which is hosted in the labs host network and uses our OpenStack infrastructure just like any end users. The Nodepool schema is not related with whatever OpenStack database backend we might.
Maybe we can collocate it with one of the labs DB host supporting OpenStack since the host is in the same network segment.
I think everyone is right. It's a separate database, strictly speaking, but it should be hosted on the same server (m5-master) as the other labs db services, with the same security policy, but different passwords. Jaime should be able to set something up in a flash, when he has a flash to spare.
I am not sure how much of an issue it can be for our databases.
I had to ask for 2 reasons: usually, misc servers are not dedicated servers, which means they could be impacted or impact other issues. For example, phabricator is separate from puppet or other services due to its load.
Second, production "misc" databases have the same level of support than wiki databases: I will wake up if something is wrong with it, so I have to make sure that nothing will go wrong. That is why understanding its usage is important to me. Thanks for the extended details.
Deploying right now. Restricting access to libnodepool1001.
Access has been granted to m5-master only from labnodepool1001:
root@labnodepool1001:~$ mysql -h m5-master -u nodepool -p Enter password: Welcome to the MySQL monitor. Commands end with ; or \g. Your MySQL connection id is 2158236 Server version: 5.5.5-10.0.16-MariaDB-log Source distribution Copyright (c) 2000, 2015, Oracle and/or its affiliates. All rights reserved. Oracle is a registered trademark of Oracle Corporation and/or its affiliates. Other names may be trademarks of their respective owners. Type 'help;' or '\h' for help. Type '\c' to clear the current input statement. mysql> SHOW DATABASES; +--------------------+ | Database | +--------------------+ | information_schema | | nodepooldb | +--------------------+ 2 rows in set (0.00 sec)
In need to add nodepooldb to the backups now.
I proposed a change upstream to amend the documentation and referencing the above link. https://review.openstack.org/#/c/219324/ has been merged and the flush privileges command disappeared from the Nodepool online documentation http://docs.openstack.org/infra/nodepool/installation.html#database
Thanks @jcrespo !