Jump to content

SRE/Production access

From Wikitech
(Redirected from Production shell access)

For instructions on accessing public Cloud VPS and Toolforge instances, see Help:Accessing Cloud VPS instances.

Production (sometimes called prod) is the network of servers that run the real, live Wikimedia websites. Access to production is necessary for deploying updates and other site reliability engineering work, as well as for accessing sensitive data. This page explains how to request and set up this access.

Remember: production access is extremely sensitive. With production access, it's possible to break our websites or steal private data about users' activities. If you have access, act carefully and take the server access responsibilities seriously. Immediately contact the SRE team if you have any doubts about security or if something goes wrong.


To minimize risk to the sites, only a small number of people outside of the SRE team hold any production access, and that access is limited to specific systems and processes. Access is managed through groups rather than people; a person (technically, an ssh-authenticated account) belongs to one or more groups, and each group has its own list of access privileges. All access privileges require a clear, ongoing need for the access. If you have a one-time need for data, request the data from the Data Engineering team instead.

There are three distinct processes for changing production access:

Change the privileges of an access group

An existing access group, usually with existing members, can be granted additional privileges, to allow members of the group to perform additional work.

  1. a ticket in Phabricator
    1. Use the tag SRE-Access-Requests.
    2. Include the name of an existing group
    3. Include the requested change in access, in as much specific detail (host names, etc) as possible.
    4. Include the reason the change is requested, including the impact if the change is rejected

Add WMF/WMDE Staff to an access group

For WMF and WMDE staff, membership in an access group is at the discretion of their manager, who should request access on behalf of the person as detailed below.

Add a volunteer to an access group

Volunteer access is granted at the discretion of the SRE team.

  1. You must have a non-disclosure agreement with the Wikimedia Foundation. Follow the volunteer NDA process.
  2. You must have support from a relevant Wikimedia Foundation employee: this should be the employee you will be collaborating with.
  3. Complete the access request process as detailed below.

Access Request Process


If you've satisified the eligibility requirements above, follow these steps to submit a request.


To follow these instructions, you'll need the following accounts:

Signing the agreement

Next, read and sign the Acknowledgement of Wikimedia Server Access Responsibilities. Make sure you actually read it; this is a legal agreement and by signing it, you are committing to follow the security practices it describes.

Generating your SSH key

Since production access uses the Secure Shell protocol (SSH), you'll have to generate a new SSH keypair. Do not reuse an existing key; this presents an unacceptable security risk.

GitHub has a good help page (note that you can switch between Mac, Windows, and Linux documentation right under the title).

We recommend that you use an ED25519 key (or, alternatively, a 4096-bit RSA key). Do not use DSA keys as they are insecure and rejected by our SSH servers.

To generate an ED25519 key, run the following command in your terminal:

ssh-keygen -t ed25519

To generate an RSA key, run the following command in your terminal:

ssh-keygen -t rsa -b 4096 -o

The newer -o option saves private keys in a slightly more secure format (OpenSSH rather than PEM)

Reminder: the key you use for production access must be different from the key you use for Cloud VPS, so do not paste it into the Openstack field under Special:Preferences on this wiki).

Filing the request

  1. Create a ticket requesting access.[1]
    1. In the title, replace "RESOURCE" and "USER" with your name and the resource you need access to. (For new user requests, make a separate ticket for each user.)
    2. Add the following information to the description:
      • USER's full name
      • USER's wikitech username
      • Your developer access username (that is, the one you use for Cloud VPS SSH, not Wikitech login. Wikitech shows this as "instance shell account name" in preferences). We will use this as your production shell username.
      • The public key from new your SSH keypair.[2] (See for example the Gerrit instructions how to copy your public SSH key.)
      • Requested group membership. A complete list of groups that USER should be added to. These groups change frequently, so consult the most recent available list where possible.
      • A detailed reason for your request. In particular, describe which specific servers you need access to and why. We err on the side of giving fewer permissions rather than more, so the more detailed your request, the more likely you are to get all the permissions you need.
  2. Get approvals from the following people as comments to the Phabricator task. The comments should be made directly through the web interface, not via email.[3]
    • The relevant Wikimedia Foundation/Wikimedia Deutschland employee, as explained above.
    • The project lead where your access will be granted. For Analytics and Data systems this is Andrew Otto.
  3. Wait for SRE approval, if needed:
    • An SRE may ask for you to validate your public SSH key "off band", meaning via a direct communication outside of Phabricator.
    • If you're requesting the same level of access as the rest of your team already has (e.g. because you've joined the team and you're requesting to be added to the group) then no further approval is necessary; your manager's on-ticket approval (or for non-staff, your WMDE's manager or WMF sponsor's on-ticket approval) is sufficient.
    • Otherwise, if you request any new level of sudo privileges for a group (or for yourself individually, outside of your group membership), then your request must have a security review at a biweekly SRE meeting. Sudo access is granted on an extremely limited basis, and will typically apply to the smallest permissions possible (user/process restricted over all). Expect this process to take at least two business weeks.
  4. When your request is approved, you will be asked to provide your full legal name, preferred email address for contact, and physical address to the Wikimedia Foundation Legal team (or your employee contact may forward this information on your behalf). This information will be used to customize a non-disclosure agreement, which you will be asked to read, comprehend, and electrically sign through the Foundation's contract management system. The agreement will be similar to the Volunteer NDA.
  5. The Wikimedia Foundation employee that will be supervising your work will coordinate final sign off by an Executive level staff of the Wikimedia Foundation when all other criteria have been met before your access is granted.
  6. Shell access and access to private data are different things. Access to data is granted to volunteers only if they have a formal collaboration with the research team.

If five business days pass without visible progress, please comment on the ticket to request an update, or directly contact the SRE on Clinic Duty that week.

Technical details

Production shell users, their keys, and their permissions are managed in modules/admin/data/data.yaml in the operations/puppet.git repository.

Setting up your access

Setting up your SSH config

The standard configuration for people not having root access is to have the ssh connection to be established on a bastion and proxy the command to the target host inside the cluster. To do this, add the following to your SSH config file (usually located at $HOME/.ssh/config), but change YOURUSERNAME to be your shell username on the Wikimedia servers:

# Turn CanonicalizeHostname on for Match to work below.
CanonicalizeHostname yes

# Defaults for all Wikimedia Foundation hosts.
Match host=*.wikimedia.org,*.wmnet
    ForwardAgent no
    IdentitiesOnly yes
    KbdInteractiveAuthentication no
    PasswordAuthentication no

# Configure the initial connection to the bastion host, with the one
# HostName closest to you.
Host bast
    HostName bast1003.wikimedia.org
    IdentityFile ~/.ssh/prod.key
    # In theory this User line shouldn't be necessary due to the Match above,
    # but in practice it seems to be.  In any case, it doesn't hurt.

# Proxy all connections to internal servers through the bastion host.
Host *.wmnet *.wikimedia.org !gerrit.wikimedia.org !bast*.wikimedia.org !gitlab.wikimedia.org
    ProxyJump bast
    IdentityFile ~/.ssh/prod.key

# Configure direct connection to the bastion hosts.
Host bast*.wikimedia.org
    IdentityFile ~/.ssh/prod.key

Host gerrit.wikimedia.org
    Port 29418
    IdentityFile ~/.ssh/cloud.key

In the example above you may replace bast1003.wikimedia.org with the bastion that is physically closest to you:

Advanced: operations config

If you will be setting up new servers or doing other administration work, you can use the below advanced configuration instead. Otherwise, skip this section. If you're not sure, you almost certainly don't need this!

Known host files

To ensure the validity of the hosts you connect to, enable the StrictHostKeyChecking yes option and create a local list of known hosts. A utility script is available to generate that list and keep it up to date. Read the instructions in the script's header for help on usage. If you need any additional help, contact the script's authors.

Before you can use the script, you'll need to bootstrap this setup with at least one bastion host. Disable strict host key checking, ssh to a bastion, and make sure the fingerprint matches what's listed at Help:SSH Fingerprints.


Do not use SSH agent forwarding (the -A command line option). Agent forwarding does not make it possible to steal your private key itself, but it does make it possible for someone to hijack your SSH agent and thus your identity, so we do not do it. The -a option (with a lower case "a") disables agent forwarding, and is thus included in the sample configurations above.

Do not use your production cluster SSH key for any other service, including Gerrit or Cloud VPS.

Other tips


If your production access has been approved but you aren't able to log in, you can ask for help in the Phabricator ticket for your access request. If you got access a long time ago and it's a new problem, you can file a new ticket and tag it with #sre.

Wherever you ask for help, make sure you include your SSH configuration (but not your key itself!) and the output you get when you run your ssh command with the -v option (verbose mode).

If you are prompted for a password when attempting to SSH into production, it generally means that your client is misconfigured -- most often you are presenting the wrong public key to the server. ssh -v can help you debug this. When debugging, in order to keep things clear, it's best to attempt to connect directly to a bastion host, e.g. ssh -v bast1002.eqiad.wmnet.

See also


  1. The form automatically adds the ticket to the SRE-Access-Requests project so the SRE team will see your request.
  2. You can also put your public key on your wiki user page, in a Phabricator paste, or in a Gerrit patchset you upload, but you can't include it in an email reply to the task.
  3. This protects against email spoofing.