draft a tools policy

tsa/policy/tpa-rfc-3-tools.mdwn

+[[!meta title="TPA-RFC-2: tools"]]
+
+Summary: we try to restrict the number of tools users and sysadmins
+need to learn to operate in our environment. This policy documents
+which tools we use.
+
+# Background
+
+A proliferation of tools can easily creep up into an organisation. By
+limiting the number of tools in use, we can keep training and
+documentation to a more reasonable size. There's also the off chance
+that someone might already know *all* or a large proportion the tools
+currently in use if the set is smaller and standard.
+
+# Proposal
+
+This proposal formally defines which tools are used and offered by TPA
+for various things inside of TPO.
+
+We try to have one and only one tool for certain services, but
+sometimes we have many. In that case, we try to deprecate one of the
+tools in favor of the other.
+
+## Scope
+
+This applies to services provided by TPA, but not necessarily to all
+services available inside TPO. Service admins, for example, might make
+different decisions than the ones described here for practical
+reasons.
+
+## Tools list
+
+This list consists of the known policies we currently have
+established.
+
+ 1. version control: git, gitolite, gitlab
+ 2. operating system: Debian packages (official, backports,
+    third-party and TPA)
+ 3. host installation: debootstrap, FAI, Fabric
+ 4. ad-hoc tools: SSH, Cumin, Python/Fabric
+ 5. directory servers: OpenLDAP, BIND, ud-ldap, Hiera
+ 6. authentication servers: OpenLDAP, ud-ldap
+ 7. time synchronisation: NTP (`ntp` Debian package, from ntp.org)
+ 8. Network File Servers: DRBD
+ 9. File Replication Servers: static mirror system
+ 10. Client File Access: N/A
+ 12. Client OS Update: unattended-upgrades, needrestart, dsa nagios
+     checks, multiple reboot tools ([ticket #33406](https://trac.torproject.org/projects/tor/ticket/33406))
+ 13. Client Configuration Management: Puppet
+ 14. Client Application Management: Debian packages, Ansible, systemd
+     lingering, cron `@reboot` targets (deprecated)
+ 15. Mail: SMTP/Postfix, Mailman, ud-ldap, dovecot (on gitlab)
+ 16. Printing: N/A
+ 17. Monitoring: syslog-ng central host, no paging, Nagios,
+     Prometheus, Grafana
+ 18. password management: pwstore
+ 19. help desk: Trac (deprecated), GitLab, RT, email, IRC
+ 20. backup services: bacula, postgresql hot sync
+ 21. web services: Apache, Nginx, Varnish (deprecated), haproxy
+     (deprecated)
+ 22. documentation: ikiwiki, Trac wiki
+ 23. datacenters: Hetzner cloud, Hetzner robot, Cymru, Sunet, Linaro,
+     Scaleway (deprecated)
+
+Changes from current informal policy:
+
+ 1. addition of Python/Fabric to adhoc tools and installer
+ 2. addition gitlab to version control
+
+## TODO
+
+ 1. figure out scope... list has grown big already
+ 2. are server specs part of this list?
+ 3. software raid?
+
+## Examples
+
+ * all changes to servers should be performed through Puppet, as much
+   as possible...
+ * ... except for services not managed by TPA ("service admin stuff"),
+   which can be deployed by hand, Ansible, or any other tool
+ * 
+
+# Deadline
+
+No deadline set yet, still drafting.
+
+# Status
+
+This proposal is currently in the `draft` state.
+
+# References
+
+Drafting this policy was inspired by the [limiting tool dev
+choices](https://utcc.utoronto.ca/~cks/space/blog/sysadmin/LimitingToolDevChoices) blog post from [Chris Siebenmann](https://utcc.utoronto.ca/~cks/) from the [University
+of Toronto Computer Science department](https://www.cs.toronto.edu/).
+
+The tool classification is a variation of the [infastructures.org](http://www.infrastructures.org/)
+[checklist](http://www.infrastructures.org/bootstrap/checklist.shtml), with item 2 changed from "Gold Server" to "Operating
+System". The naming change is rather dubious, but I felt that "Gold
+Server" didn't really apply anymore in the context of configuration
+management tools like Puppet (which is documented in item 13). Debian
+is a fundamental tool at Tor and it feels critical to put it first and
+ahead of everything else, because it's one thing that we rely on
+heavily. It also *does* somewhat acts as a "Gold Server" in that it's
+a static repository of binary code. We also do not have uniform
+"Client file access" (item 10) and "Printing" (item 16). Item 18
+("Password management") was also added.