Yaybu 3.0

Server deployment and configuration management in Python

Yaybu is a push based configuration management tool written in Python with the goal of helping you tame your servers. You describe your infrastructure in a simple and flexible YAML-like language and Yaybu works out what needs to happen to deploy your updates.

All the documentation is available on http://docs.yaybu.com.

Yaybu is very new software, please subscribe to the yaybu-dev mailing list and let us know if you are using it, we’d love to hear from you. You can find the mailing list at http://lists.yaybu.com.

You can also find us in #yaybu on irc.oftc.net.

Changelog for yaybu

3.0 (2013-09-04)

• New line of development based upon the latest iteration of yay (yay3).

• Refactor provisioner backend to use “transports” (rather than starting a remote copy of yaybu that communicates with the local yaybu)

• Refactor the libcloud code to integrate with the yay3 “graph” - so that nodes are started on demand and ordering is implicit.

0.2.4 (unreleased)

• Add a subclass of yay.config.Config that helps adapt yay for use in Yaybu..

• Depend on at least 0.11.4 of libcloud to avoid hanging on S3 access.

• Drop dependency on boto.

0.2.3 (2012-10-25)

• Update workaround to avoid conflicts with boto (which has a HTTPResponse subclass).

• Remove some of the more pointless logging.

0.2.2 (2012-10-24)

• Workaround a bug in python2.6 that causes certain Amazon S3 operations to hang.

0.2.1 (2012-10-23)

• Fixed MANIFEST to include recipes

0.2.0 (2012-10-23)

• Add new commands for dealing with collections of nodes on cloud providers: addnode, rmnode, status and provision.

• NB: 0.2.0 is API incompatible with earlier versions

• Mark “static” file content retrieved over a gpg stream as secret to prevent the content appearing in the logs.

• Depend on yay >= 0.0.57 for latest bug fixes.

• Add a new yaybu bootstrap to simplify setup on Ubuntu boxes. If you are running a development build of Yaybu then the bootstrap will be of the latest code pushed to master on GitHub.

0.1.28 (2012-10-13)

• Add a new Checkout provider for use with Vagrant-like environments:

  resources.append:
    - Checkout:
        name: /vagrant
        scm: mount

  This pattern allows recipes to treat shared folders in VM’s (like /vagrant) as checkouts. They can then be upgraded/substituted for git or subversion checkouts when deploying to a real server.

  Any files that are watched in the checkout will be treated as though they have changed.

0.1.27 (2012-08-14)

• Fix yaybu.core.shell bug where the communicate implementation waited for file handles to disappear even when the underlying process had terminated.

0.1.26 (2012-08-05)

• Restore previous verbosity (will review properly during 0.2.x).

0.1.25 (2012-08-02)

• Fix default verbosity to not be debug mode

• Use context log features of latest yay version to provide better error messages when yay parsing fails. These are only shown when verbosity is at debug level.

0.1.24 (2012-07-30)

• Fix etag request headers so that etags can be back from a yaybu running in remote mode.

0.1.23 (2012-07-30)

• Fix etag registry to actually store etags

0.1.22 (2012-07-22)

• Now depends on yay >= 0.0.53 to support etags (which reduces the amount of network traffic when files haven’t changed) and to support fetching of cookbooks packaged as python eggs from code repositories.

• Fix logging typos

• If there is a `~/.yaybu/defaults.yay` or `~/.yaybu/defaults.yay.gpg` it will be loaded.

0.1.21 (2012-03-16)

• Fix simulate error

0.1.20 (2012-03-16)

• Generate proper exceptions when trying to lookup users and groups that don’t exist.

• Refactor ‘passthru’ commands as ‘inert’ commands as it better reflects the intention.

• Remove ‘exceptions’ option from shell.execute

• #63 Simulate fails on Execute when user missing from /etc/passwd. Code refactored to work around this.

• Experimental Paramiko support.

0.1.19 (2012-03-13)

• Improved validation for scripts in cwd.

0.1.18 (2012-03-13)

• #66, fix ordering of permission dropping so that we can actually drop group permissions.

• #65, add a umask option to Execute.

• #64, proper error handling when assigning to an invalid property.

• After parsing a shell command, verify that the command exists (searching on PATH if needed).

• Make sure errors from Yay are wrapped in ParseError so they are presented properly.

0.1.17 (2012-03-12)

• Support a running option for Service. This is a command that should return an exit code of 0 if all services are running. If used with a buildout generated isotoma.recipe.cluster symlinked into /etc/init.d this might look like this:

  resources.append:
    - Service:
        name: cluster
        running: /etc/init.d/cluster running

0.1.16 (2012-01-16)

• Fix handling of broken links when replacing them.

• Fix handling of broken links for Link remove policy.

0.1.15 (2012-01-08)

• If remote process dies before we SIGKILL it, swallow any errors.

• The RemoteRunner object can now set SSH settings in its ssh_options dictionary. For example:

  runner.ssh_options["IdentitiesOnly"] = "yes"

0.1.14 (2012-01-01)

• Better error handling. Proper detection of resource redefinition and of undefined resource type use. Yay node graph is walked lazily so Yaybu errors can access line numbers and other metadata. The resource bundle is parsed and error checked before we even try and connect to server.

• Exit code 254 is now used when no changes have been made. 253 is used for unknown errors. 255 is now a connection errror.

• Refactor template handling to allow {% extends "index.html" %}. You can refer to templates on the YAYBUPATH, and you can override templates by putting new templates higher up the YAYBUPATH search.

• Yaybu hackers now have additional ways of invoking the debugger, including an equivalent to the pdb pm() and set_trace() methods..

• File creation is now more secure: permissions are set before content is written.

• We now use Yay 0.0.41 as it has a cleaner API for injecting Python objects.

0.1.13 (2011-12-02)

• Initial subversion checkouts are no longer quiet - to match ‘svn up’ behaviour - this is useful for large checkouts when you arent sure if the checkout has hung.

• Use of ‘svn’ without having subversion installed will trigger a propper error instead of a traceback. Both this and the rsync checks now have a proper exception for reporting missing dependencies.

• Service[] now parses the output of /sbin/status to determine if an upstart job is running or not. It considers the goal of the job to determine if it needs to adjust the job. For example, we might want to start a job and that job isn’t running, but it is starting. That is good enough, so Yaybu wouldn’t take any action.

  A side effect of implementing proper state detection is that when trying to restart an upstart job that was stopped now just starts the job directly.

• Adds version information to the command line interface

• The protocol Yaybu uses for communicating over SSH has grown some version metadata. This will be used to ensure better interoperability between Yaybu clients and targets.

  WARNING You will need to ensure your clients and servers are running at least this version as this is (hopefully the last) protcol breaking change.

• Reclassifies some log output as debug information for quieter deployments when nothing has changed.

0.1.12 (2011-11-30)

• Add a dummy Checkout provider that uses rsync. The primary use case for this feature is vagrant where you can do something like this:

  resources.append:
    - Checkout:
        name: /var/checkouts/my-site
        repository: /vagrant
        scm: rsync
        user: vagrant
  
    - Execute:
        name: buildout
        command: ./buildout
        cwd: /var/checkouts/my-site
        policy:
          execute:
            - when: sync
              on: Checkout[/var/checkouts/my-site]

  There are 2 advantages here: firstly, your production config will likely deploy from Git, but Vagrant always gives you a /vagrant directory without having to have a seperate Git checkout. With this change you can still use a Checkout resource for your /vagrant sourced code, keeping your recipes as similar as possible. Secondly, your vagrant provision will be able to detect code changes just like a real Checkout. For example, you can add watches to restart services when you change a particular script or asset in your ‘checkout’.

• The default mode for directories is now 0755.

• Improve cross-platform support by not making assumptions about the value of sys.maxunicode.

• Use yay 0.0.39 to avoid some pickle issues.

0.1.11 (2011-11-19)

• A Subversion checkout will now understand and happily use data passed to it as String objects. This means if you use the scm_password attribute it won’t leak your password into the logs.

• We now depend on Yay 0.0.35 for the improved String API.

• The subprocess abstraction natively supports String instances and will mask any secrets in log output. It now also supports native strings and will shlex.split them automatically.

0.1.10 (2011-11-10)

• Yabyu now depends on Yay 0.0.33 and the improved Openers() API. This means that cookbook assets can be fetched from any location that Yay can be fetched from, including http and https.

• This change also means that Yay loading also respects the YAYBUPATH, and that http content can be on the YAYBUPATH.

• Fix python-magic fallback - in previous version the fallback was inverted and claiming all text was binary

0.1.9 (2011-11-07)

• Add a basic docs for Vagrant provisioner.

• Add a better fallback for when ‘python-magic’ isn’t installed

0.1.8 (2011-11-04)

• Allow set_interactive call on RemoteRunner for batch processing use cases

• Allow API users to set a particular identity file with set_identity_file

• Change Context object to parse user@host:port into individual variables This fixes ${yaybu.host} containing user and ports details.

0.1.7 (2011-10-18)

• Add some toys for people driving Yaybu from code without Yay

• Improve quick start documentation to stress idempotence

• Don’t import yaybu.resources or yaybu.providers from yaybu package any more

• yaybu is now a namespace package

• You can declare your own resources and providers using the yaybu.resources yaybu.providers entry points

0.1.6 (2011-07-26)

• A variable called yaybu.host is available from your configs

• Add Git support

• Added minimal RemoteRunner API

• –expand-only only shows resources key (more useful)

• More test coverage

• Documentation updates

• Support ‘.import: - ${foo}’

0.1.5 (2011-07-06)

• Logging is now via python logging module

• Audit log to syslog is working again

• Audit log is configurable through /etc/yaybu

• Added yaybu.core.debug to assist debugging yaybu in remote mode

• Can run network tests with TEST_NETWORK=1 ./bin/test discover

• New yaybu.harness module to house testing harness related code

• Tests have moved into sensible locations within yaybu module

0.1.4 (2011-06-16)

• EventState removes successfully applied updates on the fly, not just at the end

• Remove the simlog API and be as ‘full-fat’ as possible in simulate mode

• File provider will no longer log diffs when updating encrypted files ;)

• Every test that uses check_apply will also be tested for simulation

• Better use of python-magic: Use mimetypes and look for text/

• Use a stricter glob for identifying currently active sysv services

• EventState now records against id rather than home

• Test harness now enforces idempotence of check_apply

• Test harness now manages fakeroot by itself

• Fix logging issue for long resource ids

• More sensible default PATH

• Don’t consider empty files to be binary

• Simplify logging, and fix logging of diffs

• Set environment from preexec function

• Lock down environment - don’t leak environment from user

• Be more forgiving of mising PID files (Service provider)

• Better error handling in remote mode

0.1.3 (2011-05-26)

• RemoteRunner fixes

• Fix catching errors in remote

• Fix passing –resume and –no-resume to remote

0.1.2 (2011-05-22)

• Basic upstart, debian and ‘simple’ service providers

• Catches yay.errors.Error

• Add ‘live logging’ of shell steps (so you can see buildout progress)

• Add ability to watch files that might be updated by other providerss

• More tests

• Test env deals with races from cow-shell automatically

• Fix comment field of User provider

• Fix group and user bit of File/Directory/etc

• Resources are now referred to as Type[name] as otherwise Group and User can very easily crash

• Jinja use is now more tolerant of unicode dictionary keys

• Observation state is now preserved on failure