You are browsing a read-only backup copy of Wikitech. The primary site can be found at wikitech.wikimedia.org

Difference between revisions of "Homer"

From Wikitech-static
Jump to navigation Jump to search
imported>Ayounsi
imported>Ayounsi
Line 156: Line 156:
</syntaxhighlight>
</syntaxhighlight>
*Can't have jinja applied to it, so filters like ping-offload needs to stay out of Capirca
*Can't have jinja applied to it, so filters like ping-offload needs to stay out of Capirca
*POC limitation: Netbox definitions need to be manually updated by running https://netbox-next.wikimedia.org/extras/scripts/capirca.GetHosts/
*  
*  


Line 241: Line 242:
* '''Error parsing cr-analytics: UNDEFINED: puppetmaster.'''
* '''Error parsing cr-analytics: UNDEFINED: puppetmaster.'''
** There is a <code>{source|destination}-address:: puppetmaster</code> in cr-analytics.pol (or one of its child includes) not defined static.net or Netbox
** There is a <code>{source|destination}-address:: puppetmaster</code> in cr-analytics.pol (or one of its child includes) not defined static.net or Netbox
**See https://netbox.wikimedia.org/extras/scripts/#script.GetHosts "Last run" then "output" tab to see the Netbox generated definitions.
* '''Error parsing cr:  ERROR on "udp" (type STRING, line 38, Next 'destination-port').'''  '''Error parsing cr-analytics:  ERROR on "T274951" (type STRING, line 312, Next 'destination-address').'''
* '''Error parsing cr:  ERROR on "udp" (type STRING, line 38, Next 'destination-port').'''  '''Error parsing cr-analytics:  ERROR on "T274951" (type STRING, line 312, Next 'destination-address').'''
** Most common cause is forgetting the double semi-colon "::" or forgetting the quotes around a comment
** Most common cause is forgetting the double semi-colon "::" or forgetting the quotes around a comment

Revision as of 18:08, 12 March 2021

Homer (previously jnt) is our homemade network configuration manager.

It takes variables from Netbox and yaml files, run them through jinja templates to generate Juniper compatible configuration.

Homer can then send those configurations to selected network devices, for a diff or a safe commit.

The tool is written to not be Wikimedia specific. It only supports Junos but can easily be extended to other platforms.

Its doc is available on https://doc.wikimedia.org/homer/master/

Its code on Gerrit https://gerrit.wikimedia.org/g/operations/software/homer

Its bug and feature requests on Phabricator: https://phabricator.wikimedia.org/tag/homer/

This page focuses on Wikimedia's deployment.

Deployment

Homer is deployed via Puppet and Scap to the cumin (fleet management) hosts.

You can find its deploy repository here https://gerrit.wikimedia.org/g/operations/software/homer/deploy

And its Puppet module there https://gerrit.wikimedia.org/r/plugins/gitiles/operations/puppet/+/production/modules/homer

In addition it's available on Pypi: https://pypi.org/project/homer/

Releasing a new version

  • Make a release patch updating the changelog (see this example patch).
  • Once it's merged, update the local checkout and make a git tag. Ideally an annotated one (requires a GPG key and have git configured to use it, see signingkey):
$ RELEASE=v0.1.0
$ git tag -s -a "${RELEASE}" -m "${RELEASE}" -m "[Release Notes](CHANGELOG.rst)"
  • Push the generated tag: git push origin "${RELEASE}"
  • Move to the homer-deploy checkout:
$ cd src/
$ git pull
$ git log -1  # to check to be at the right commit
$ cd ..
# At this point git status would show that there is a diff for the 'src' path, indicating the different SHA1 of the git submodule
# Ensure that docker is running
$ make -f Makefile.build all
# Verify that the generated wheels are correct
# At this point the frozen-requirements.txt file will most likely have some changes and the artifacts/artifacts.stretch.tar.gz will be different
git add .
git commit -m "Release ${RELEASE}"
git review
  • Once the above patch has been merged (C+2, V+2 + submit), move to the deployment server in /srv/deployment/homer/deploy
  • Pull the latest changes: git pull
  • Verify that in the src/ directory we're at the correct commit (check also with git status)
  • Deploy the new release: scap deploy --verbose "Homer release v... - T..."

Daily diffs

A cron job runs Homer every 12h (24h per cumin hosts) to compare the live network configuration with our intended state. Any discrepancies is emailed to the rancid-core alias.

Usage 🚀

Making changes

Note that Homer explicitly asks you when its about to modify the live network configuration (Type "yes" to commit, "no" to abort.) and will prompt you with a diff of the changes beforehand.

Editing the private repository

Manually edit then commit the files on ssh://cumin1001.eqiad.wmnet:/srv/homer/private .

git will sync them with the other cumin host. And will email a summary of the changes to SREs.

Make sure to mirror all your changes on the mock-private repo: https://gerrit.wikimedia.org/g/operations/homer/mock-private

This repository doesn't have CI, please be extra careful.

Editing the public repository

Similar to our other public repositories, send CRs to https://gerrit.wikimedia.org/g/operations/homer/public , try not to self-+2 your changes without other review.

Editing Netbox

Data is also pulled from Netbox, always make sure that Netbox is accurate before using Homer.

Running Homer from cumin hosts (recommended)

Get familiar with the command line: https://doc.wikimedia.org/homer/master/homer.html everything else is taken care of.

The public repository is regularly updated by Puppet.

When pushing configurations, homer will ssh to the network devices using the Homer user. You need to be in the ops group to be able to use its private key.

Some examples:

  • homer "*" diff All devices
  • homer "cr*ams*" diff esams and knams core routers
  • homer "mr*" commit "My commit message" All management routers

Running Homer from your local machine (less recommended)

When pushing configurations, your machine will ssh directly to the network devices, which mean that you have to have an account there, with the proper permissions.

It's common to test a change locally with the "diff" option. Once satisfied with the result, please merge your change on Gerrit before pushing them with the "commit" action.

Style guides

YAML files

Set configuration options doc as comment in the most common denominator file, with a note if it can be applied to more specific sites/roles/devices. See existing options.

TODO

https://json-schema.org

Templates

It's ok to give up on indentation.

Capirca (ACL generation)

This is still a POC.

Task: https://phabricator.wikimedia.org/T273865

Capirca is an actively maintained open source tool made by Google to generate multi-platform ACLs based on generic policy and definitions files.

How it works?

File:Capirca.png

  1. User edits relevant files (see bellow)
  2. User run Homer
  3. Homer pulls the hosts definitions from Netbox
  4. Homer executes Capirca for each relevant policy files (defined in homer-{public|private}/config/{devices|roles}.yaml )
  5. Capirca takes all the (hosts/services) definition files as input, as well as the policy files (while following the includes) and generates the firewall rules in the proper format
  6. Homer adds the previously generated file to the other parts of the generated config and pushes it to the device

Advantages

  • IPv4 and IPv6 filters will be updated automatically as long as the hosts have both v4 and v6 records
  • Limited blast radius if a mistake is done in a given .inc policy file
  • Centralized services (ports) definitions in text file
  • Hosts definitions synced up from Netbox
  • Same syntax for all platforms (Juniper and JuniperSRX in our case)
  • Shading detection (eg. useless rules hidden behind a more generic one)
  • Reduced operational complexity
  • Easier to audit

Limitations

  • Dependency on Netbox
  • Increased setup complexity
  • Doesn't distinguish between v4 and v6 prefix-lists, hat means to leverage the auto-generation of both v4 and v6 we have to specify both prefix-lists
    [edit firewall family inet filter loopback4 term return-tcp from source-prefix-list]
             wikimedia4 { ... }
    +        wikimedia6;
    
  • Can't have jinja applied to it, so filters like ping-offload needs to stay out of Capirca
  • POC limitation: Netbox definitions need to be manually updated by running https://netbox-next.wikimedia.org/extras/scripts/capirca.GetHosts/

How to use it?

Update an existing ACL

  1. Browse homer-{public|private}/policies
  2. Find the relevant .pol or .inc file (eg. cr-analytics.pol)
  3. Update it (use existing rules and guidelines as models)
Guidelines
term my-term {
  comment:: "T123456"  # Don't forget the quotes
  destination-address:: foo # from either static.net or Netbox
  destination-port:: bar  # From the services.svc file
  action:: deny
}

term allow_rest {
  action:: accept # All our platforms have a default deny
}
  • {source|destination}-port:: are defined in in the homer-public/definitions/services.svc file, add yours if it's not already there. Ordered by port numbers.
  • Most {source|destination}-address:: are pulled from Netbox and grouped by their hostname prefix (eg. all "aqs*" hosts are under "aqs_group")
    • For network prefixes and special IPs (eg. VIPs), add them to homer-public/definitions/static.net

Add a new ACL (firewall filter)

Most likely for a Netops.

  1. Create a .pol file in homer-{public|private}/policies (eg. my-filter.pol), see guidelines bellow
  2. Reference the above policy file in either:
    • homer-{public|private}/config/{devices|roles}.yaml (recommended)
      capirca:
          - my-filter  # The policy file name without the extention
      
    • Another .pol file with #include 'my-filter.pol'
Guidelines
  • Example juniper headers
    header {
      comment:: "foobar"
      target:: juniper my-filter4 inet
      target:: juniper my-filter6 inet6
    }
    # juniper: platform (which final syntax to use)
    # my-filter4: the juniper filter name that will be generated
    # inet: IP family to target (ipv4 s. ipv6)
  • Example SRX security policies headers
    header {
        comment:: "Generated by Capirca"
        target:: srx from-zone production to-zone production address-book-global
    }
    # srx: platform
    # from/to security-zones (they need to already exist)
    # Use global address-book (default everywhere in our infra)
  • If you have to have different terms for v4 and v6 policies, put all the common policies in a .inc file, then include it before/after the specific term. For example:
    header {
      target:: juniper border-in4 inet
    }
    term offload-ping {
        verbatim:: juniper "term offload-ping4 {"
        verbatim:: juniper "    filter offload-ping4;"
        verbatim:: juniper "}"
    }
    #include 'cr-border-in.inc'
    
    header {
      target:: juniper border-in6 inet6
    }
    #include 'cr-border-in.inc'
  • If a specific Juniper syntax is not supported by Capirca, use the verbatim:: keyword, that will be copied as-it

Common errors

  • Error parsing cr: No such service, foo.
    • There is a {source|destination}-port:: foo in cr.pol (or one of its child includes) not defined in services.svc
  • Error parsing cr-analytics: UNDEFINED: puppetmaster.
  • Error parsing cr:  ERROR on "udp" (type STRING, line 38, Next 'destination-port'). Error parsing cr-analytics:  ERROR on "T274951" (type STRING, line 312, Next 'destination-address').
    • Most common cause is forgetting the double semi-colon "::" or forgetting the quotes around a comment
    • Note that it shows the next line, an the lines don't always match if there are includes.
  • Multiple definitions found for service:  git-ssh.eqiad
    • The service is defined twice, either in services.svc or in network definitions (static.net or Netbox)

References

Capirca policy format (which keywords are accepted?)

Network configuration coverage

CR

Done

groups {}
apply-groups [ re0 re1 ];
system {}
interfaces {}
logical-systems {}
services {}
snmp {}
forwarding-options {}
routing-options {}
protocols {
    ospf {}
    ospf3 {}
    lldp {}
    bgp {} # Transit, LVS, Netflow ,k8s, Anycast
}
policy-options {}
firewall {}
routing-instances {}
chassis {} (partial)

TODO

chassis {} (partial)
routing-options {}  # TODO: statics
protocols {
    router-advertisement {}
    bgp {}  # TODO: confed. IXPs are out of scope (dedicated tool like peering-manager)
}

ASW

Done

system {}
interfaces {}
snmp {}
protocols {}
routing-options {}
virtual-chassis {}
vlans {}
chassis {}

MR

Done

groups {}
system {}
snmp {}
interfaces {}
protocols {}
routing-options {}
policy-options {}
security {
    zones {}
    alg {}
    forwarding-options {}
    screen {}
    nat {}
}

TODO

security {
    address-book {}  # Capirca?
    policies {}  # Capirca?
}
routing-instances {}
applications {}  # Capirca?

MSW (all done)

system {}
interfaces {}
snmp {}
protocols {}
routing-options {}
vlans {}

CLOUDSW

TODO

Common/known issues

(Almost) None.

  • The "commit" action will not work on the first try with the mr1* devices, but homer will retry.