diff --git a/CHANGELOG b/CHANGELOG index 290a8c94f..0908aeaa5 100644 --- a/CHANGELOG +++ b/CHANGELOG @@ -1,913 +1,916 @@ + Fixed #971 -- classes can once again be included multiple + times. + Added builtin support for Nagios types using Naginator to parse and generate the files. 0.24.1 Updated vim filetype detection. (#900 and #963) Default resources like schedules no longer conflict with managed resources. (#965) Removing the ability to disable http keep-alive, since it didn't really work anyway and it should no longer be necessary. Refactored http keep-alive so it actually works again. This should be sufficient enough that we no longer need the ability to disable keep-alive. There is now a central module responsible for managing HTTP instances, along with all certificates in those instances. Fixed a backward compatibility issue when running 0.23.x clients against 0.24.0 servers -- relationships would consistently not work. (#967) Closing existing http connections when opening a new one, and closing all connections after each run. (#961) Removed warning about deprecated explicit plugins mounts. 0.24.0 (misspiggy) Modifying the behaviour of the certdnsnames setting. It now defaults to an empty string, and will only be used if it is set to something else. If it is set, then the host's FQDN will also be added as an alias. The default behaviour is now to add 'puppet' and 'puppet.$domain' as DNS aliases when the name for the cert being signed is equal to the signing machine's name, which will only be the case for CA servers. This should result in servers always having the alias set up and no one else, but you can still override the aliases if you want. External node support now requires that you set the 'node_terminus' setting to 'exec'. See the IndirectionReference on the wiki for more information. http_enable_post_connection_check added as a configuration option for puppetd. This defaults to true, which validates the server SSL certificate against the requested host name in new versions of ruby. See #896 for more information. Mounts no longer remount swap filesystems. Slightly modifying how services manage their list of paths (and adding documention for it). Services now default to the paths specified by the provider classes. Removed 'type' as a valid attribute for services, since it's been deprecated since the creation of providers. Removed 'running' as a valid attribute for services, since it's been deprecated since February 2006. Added modified patch by Matt Palmer which adds a 'plugins' mount, fixing #891. See PluginsInModules on the wiki for information on usage. Empty dbserver and dbpassword settings will now be ignored when initializing Rails connections (patch by womble). Configuration settings can now be blank (patch by womble). Added calls to endpwent/endgrent when searching for user and group IDs, which fixes #791. Obviated 'target' in interfaces, as all file paths were automatically calculated anyway. The parameter is still there, but it's not used and just generates a warning. Fixing some of the problems with interface management on Red Hat. Puppet now uses the :netmask property and does not try to set the bootproto (#762). You now must specify an environment and you are required to specify the valid environments for your site. (#911) Certificates now always specify a subjectAltName, but it defaults to '*', meaning that it doesn't require DNS names to match. You can override that behaviour by specifying a value for 'certdnsnames', which will then require that hostname as a match (#896). Relationship metaparams (:notify, :require, :subscribe, and :before) now stack when they are collecting metaparam values from their containers (#446). For instance, if a resource inside a definition has a value set for 'require', and you call the definition with 'require', the resource gets both requires, where before it would only retain its initial value. Changed the behavior of --debug to include Mongrel client debugging information. Mongrel output will be written to the terminal only, not to the puppet debug log. This should help anyone working with reverse HTTP SSL proxies. (#905) Fixed #800 -- invalid configurations are no longer cached. This was done partially by adding a relationship validation step once the entire configuration is created, but it also required the previously-mentioned changes to how the configuration retrieval process works. Removed some functionality from the Master client, since the local functionality has been replaced with the Indirector already, and rearranging how configuration retrieval is done to fix ordering and caching bugs. The node scope is now above all other scopes besides the 'main' scope, which should help make its variables visible to other classes, assuming those classes were not included in the node's parent. Replaced GRATR::Digraph with Puppet::SimpleGraph as the base class for Puppet's graphing. Functionality should be equivalent but with dramatically better performance. The --use-nodes and --no-nodes options are now obsolete. Puppet automatically detects when nodes are defined, and if they are defined it will require that a node be found, else it will not look for a node nor will it fail if it fails to find one. Fixed #832. Added the '--no-daemonize' option to puppetd and puppetmasterd. NOTE: The default behavior of 'verbose' and 'debug' no longer cause puppetd and puppetmasterd to not daemonize. Added k5login type. (#759) Fixed CA race condition. (#693) Added shortname support to config.rb and refactored addargs 0.23.2 Fixed the problem in cron jobs where environment settings tended to multiple. (#749) Collection of resources now correctly only collects exported resources again. This was broken in 0.23.0. (#731) 'gen_config' now generates a configuration with all parameters under a heading that matches the process name, rather than keeping section headings. Refactored how the parser and interpreter relate, so parsing is now effectively an atomic process (thus fixing #314 and #729). This makes the interpreter less prone to error and less prone to show the error to the clients. Note that this means that if a configuration fails to parse, then the previous, parseable configuration will be used instead, so the client will not know that the configuration failed to parse. Added support for managing interfaces, thanks to work by Paul Rose. Fixed #652, thanks to a patch by emerose; --fqdn again works with puppetd. Added an extra check to the Mongrel support so that Apache can be used with optional cert checking, instead of mandatory, thus allowing Mongrel to function as the CA. This is thanks to work done by Marcin Owsiany. 0.23.1 (beaker) You can now specify relationships to classes, which work exactly like relationships to defined types: require => Class[myclass] This works with qualified classes, too. You can now do simple queries in a collection of exported resources. You still cannot do multi-condition queries, though. (#703) puppetca now exits with a non-zero code if it cannot find any host certificates to clean. (Patch by Dean Wilson.) Fully-qualified resources can now have defaults. (#589) Resource references can now be fully-qualified names, meaning you can list definitions with a namespace as dependencies. (#468) Files modified using a FileType instance, as ParsedFile does, will now automatically get backed up to the filebucket named "puppet". Added a 'maillist' type for managing mailing lists. Added a 'mailalias' type for managing mail aliases. Added patch by Valentin Vidic that adds the '+>' syntax to resources, so parameter values can be added to. The configuration client now pulls libraries down to $libdir, and all autoloading is done from there with full support for any reloadable file, such as types and providers. (#621) Note that this is not backward compatible -- if you're using pluginsync right now, you'll need to disable it on your clients until you can upgrade them. The Rails log level can now be set via (shockingly!) the 'rails_loglevel' parameter (#710). Note that this isn't exactly the feature asked for, but I could not find a way to directly copy ActiveRecord's concept of an environment. External node sources can now return undefined classes (#687). Puppet clients now have http proxy support (#701). The parser now throws an error when a resource reference is created for an unknown type. Also, resource references look up defined types and translate their type accordingly. (#706) Hostnames can now be double quoted. Adding module autoloading (#596) -- you can now 'include' classes from modules without ever needing to specifically load them. Class names and node names now conflict (#620). 0.23.0 Modified the fileserver to cache file information, so that each file isn't being read on every connection. Also, added londo's patch from #678 to avoid reading entire files into memory. Fixed environment handling in the crontab provider (#669). Added patch by trombik in #572, supporting old-style freebsd init scripts with '.sh' endings. Added fink package provider (#642), as provided by 'do'. Marked the dpkg package provider as versionable (#647). Applied patches by trombik to fix FreeBSD ports (#624 and #628). Fixed the CA server so that it refuses to send back a certificate whose public key doesn't match the CSR. Instead, it tells the user to run 'puppetca --clean'. Invalid certificates are no longer written to disk (#578). Added a package provider (appdmg) able to install .app packages on .dmg files on OS X (#641). Applied the patch from #667 to hopefully kill the client hanging problems (permanently, this time). Fixed functions so that they accept most other rvalues as valid values (#548). COMPATIBILITY ALERT: Significantly reworked external node support, in a way that's NOT backward-compatible: Only ONE node source can be used -- you can use LDAP, code, or an external node program, but not more than one. LDAP node support has two changes: First, the "ldapattrs" attribute is now used for setting the attributes to retrieve from the server (in addition to required attriutes), and second, all retrieved attributes are set as variables in the top scope. This means you can set attributes on your LDAP nodes and they will automatically appear as variables in your configurations. External node support has been completely rewritten. These programs must now generate a YAML dump of a hash, with "classes" and "parameters" keys. The classes should be an array, and the parameters should be a hash. The external node program has no support for parent nodes -- the script must handle that on its own. Reworked the database schema used to store configurations with the storeconfigs option. Replaced the obsolete RRD ruby library with the maintained RubyRRDtool library (which requires rrdtool2) (#659). The Portage package provider now calls eix-update automatically when eix's database is absent or out of sync (#666). Mounts now correctly handle existing fstabs with no pass or dump values (#550). Mounts now default to 0 for pass and dump (#112). Added urpmi support (#592). Finishing up the type => provider interface work. Basically, package providers now return lists of provider instances. In the proces, I rewrote the interface between package types and providers, and also enabled prefetching on all packages. This should significantly speed up most package operations. Hopefully fixing the file descriptor/open port problems, with patches from Valentin Vidic. Significantly reworked the type => provider interface with respect to listing existing provider instances. The class method on both class heirarchies has been renamed to 'instances', to start. Providers are now expected to return provider instances, instead of creating resources, and the resource's 'instances' method is expected to find the matching resource, if any, and set the resource's provider appropriately. This *significantly* reduces the reliance on effectively global state (resource references in the resource classes). This global state will go away soon. Along with this change, the 'prefetch' class method on providers now accepts the list of resources for prefetching. This again reduces reliance on global state, and makes the execution path much easier to follow. Fixed #532 -- reparsing config files now longer throws an exception. Added some warnings and logs to the service type so users will be encouraged to specify either "ensure" or "enabled" and added debugging to indicate why restarting is skipped when it is. Changed the location of the classes.txt to the state directory. Added better error reporting on unmatched brackets. Moved puppetd and puppetmasterd to sbin in svn and fixed install.rb to copy them into sbin on the local system appropriately. (#323) Added a splay option (#501). It's disabled when running under --test in puppetd. The value is random but cached. It defaults to the runinterval but can be tuned with --splaylimit Changing the notify type so that it always uses the loglevel. Fixing #568 - nodes can inherit from quoted node names. Tags (and thus definitions and classes) can now be a single character. (#566) Added an 'undef' keyword (#629), which will evaluate to "" within strings but when used as a resource parameter value will cause that parameter to be evaluated as undefined. Changed the topological sort algorithm (#507) so it will always fail on cycles. Added a 'dynamicfacts' configuration option; any facts in that comma-separated list will be ignored when comparing facts to see if they have changed and thus whether a recompile is necessary. Renamed some poorly named internal variables: @models in providers are now either @resource or @resource_type (#605). @children is no longer used except by components (#606). @parent is now @resource within parameters (#607). The old variables are still set for backward compatibility. Significantly reworking configuration parsing. Executables all now look for 'puppet.conf' (#206), although they will parse the old-style configuration files if they are present, although they throw a deprecation warning. Also, file parameters (owner, mode, group) are now set on the same line as the parameter, in brackets. (#422) Added transaction summaries (available with the --summarize option), useful for getting a quick idea of what happened in a transaction. Currently only useful on the client or with the puppet interpreter. Changed the interal workings for retrieve and removed the :is attribute from Property. The retrieve methods now return the current value of the property for the system. Removed acts_as_taggable from the rails models. 0.22.4 Execs now autorequire the user they run as, as long as the user is specified by name. (#430) Files on the local machine but not on the remote server during a source copy are now purged if purge => true. (#594) Providers can now specify that some commands are optional (#585). Also, the 'command' method returns nil on missing commands, rather than throwing an error, so the presence of commands be tested. The 'useradd' provider for Users can now manage passwords. No other providers can, at this point. Parameters can now declare a dependency on specific features, and parameters that require missing features will not be instantiated. This is most useful for properties. FileParsing classes can now use instance_eval to add many methods at once to a record type. Modules no longer return directories in the list of found manifests (#588). The crontab provider now defaults to root when there is no USER set in the environment. Puppetd once again correctly responds to HUP. Added a syntax for referring to variables defined in other classes (e.g., $puppet::server). STDIN, STDOUT, STDERR are now redirected to /dev/null in service providers descending from base. Certificates are now valid starting one day before they are created, to help handle small amounts of clock skew. Files are no longer considered out of sync if some properties are out of sync but they have no properties that can create the file. 0.22.3 Fixed backward compatibility for logs and metrics from older clients. Fixed the location of the authconfig parameters so there aren't loading order issues. Enabling attribute validation on the providers that subclass 'nameservice', so we can verify that an integer is passed to UID and GID. Added a stand-alone filebucket client, named 'filebucket'. Fixed the new nested paths for filebuckets; the entire md5 sum was not being stored. Fixing #553; -M is no longer added when home directories are being managed on Red Hat. 0.22.2 (grover) Users can now manage their home directories, using the managehome parameter, partially using patches provided by Tim Stoop and Matt Palmer. (#432) Added 'ralsh' (formerly x2puppet) to the svn tree. When possible it should be added to the packages. The 'notify' type now defaults to its message being the same as its name. Reopening $stdin to read from /dev/null during execution, in hopes that init scripts will stop hanging. Changed the 'servername' fact set on the server to use the server's fqdn, instead of the short-name. Changing the location of the configuration cache. It now defaults to being in the state directory, rather than in the configuration directory. All parameter instances are stored in a single @parameters instance variable hash within resource type instances. We used to use separate hashes for each parameter type. Added the concept of provider features. Eventually these should be able to express the full range of provider functionality, but for now they can test a provider to see what methods it has set and determine what features it provides as a result. These features are integrated into the doc generation system so that you get feature documentation automatically. Switched apt/aptitide to using "apt-cache policy" instead of "apt-cache showpkg" for determining the latest available version. (#487) FileBuckets now use a deeply nested structure for storing files, so you do not end up with hundreds or thousands of files in the same directory. (#447) Facts are now cached in the state file, and when they change the configuration is always recompiled. (#519) Added 'ignoreimport' setting for use in commit hooks. This causes the parser to ignore import statements so a single file can be parse-checked. (#544) Import statements can now specify multiple comma-separated arguments. Definitions now support both 'name' and 'title', just like any other resource type. (#539) Added a generate() command, which sets values to the result of an external command. (#541) Added a file() command to read in files with no interpolation. The first found file has its content returned. puppetd now exits if no cert is present in onetime mode. (#533) The client configuration cache can be safely removed and the client will correctly realize the client is not in sync. Resources can now be freely deleted, thus fixing many problems introduced when deletion of required resources was forbidden when purging was introduced. Only resources being purged will not be deleted. Facts and plugins now download even in noop mode (#540). Resources in noop mode now log when they would have responded to an event (#542). Refactored cron support entirely. Cron now uses providers, and there is a single 'crontab' provider that handles user crontabs. While this refactor does not include providers for /etc/crontab or cron.d, it should now be straightforward to write those providers. Changed the parameter sorting so that the provider parameter comes right after name, so the provider is available when the other parameters and properties are being created. Redid some of the internals of the ParsedFile provider base class. It now passes a FileRecord around instead of a hash. Fixing a bug related to link recursion that caused link directories to always be considered out of sync. The bind address for puppetmasterd can now be specified with --bindaddress. Added (probably experimental) mongrel support. At this point you're still responsible for starting each individual process, and you have to set up a proxy in front of it. Redesigned the 'network' tree to support multiple web servers, including refactoring most of the structural code so it's much clearer and more reusable now. Set up the CA client to default to ca_server and ca_port, so you can easily run a separate CA. Supporting hosts with no domain name, thanks to a patch from Dennis Jacobfeuerborn. Added an 'ignorecache' option to tell puppetd to force a recompile, thanks to a patch by Chris McEniry. Made up2date the default for RHEL < 4 and yum the default for the rest. The yum provider now supports versions. Case statements correctly match when multiple values are provided, thanks to a patch by David Schmitt. Functions can now be called with no arguments. String escapes parse correctly in all cases now, thanks to a patch by cstorey. Subclasses again search parent classes for defaults. You can now purge apt and dpkg packages. When doing file recursion, 'ensure' only affects the top-level directory. States have been renamed to Properties. 0.22.1 (kermit) -- Mostly a bugfix release Compile times now persist between restarts of puppetd. Timeouts have been added to many parts of Puppet, reducing the likelihood if it hanging forever on broken scripts or servers. All of the documentation and recipes have been moved to the wiki by Peter Abrahamsen and Ben Kite has moved the FAQ to the wiki. Explicit relationships now override automatic relationships, allowing you to manually specify deletion order when removing resources. Resources with dependencies can now be deleted as long as all of their dependencies are also being deleted. Namespaces for both classes and definitions now work much more consistently. You should now be able to specify a class or definition with a namespace everywhere you would normally expect to be able to specify one without. Downcasing of facts can be selectively disabled. Cyclic dependency graphs are now checked for and forbidden. The netinfo mounts provider was commented out, because it really doesn't work at all. Stupid NetInfo stores mount information with the device as the key, which doesn't work with my current NetInfo code. Otherwise, lots and lots of bugfixes. Check the tickets associated with the 'kermit' milestone. 0.22.0 Integrated the GRATR graph library into Puppet, for handling resource relationships. Lots of bug-fixes (see bugs tickets associated with the 'minor' milestone). Added new 'resources' metatype, which currently only includes the ability to purge unmanaged resources. Added better ability to generate new resource objects during transactions (using 'generate' and 'eval_generate' methods). Rewrote all Rails support with a much better database design. Export/collect now works, although the database is incompatible with previous versions. Removed downcasing of facts and made most of the language case-insensitive. Added support for printing the graphs built during transactions. Reworked how paths are built for logging. Switched all providers to directly executing commands instead of going through a subshell, which removes the need to quote or escape arguments. 0.20.1 Mostly a bug-fix release, with the most important fix being the multiple-definition error. Completely rewrote the ParsedFile system; each provider is now much shorter and much more maintainable. However, fundamental problems were found with the 'port' type, so it was disabled. Also, added a NetInfo provider for 'host' and an experimental NetInfo provider for 'mount'. Made the RRDGraph report *much* better and added reference generation for reports and functions. 0.20.0 Significantly refactored the parser. Resource overrides now consistently work anywhere in a class hierarchy. The language was also modified somewhat. The previous export/collect syntax is now used for handling virtual objects, and export/collect (which is still experimental) now uses double sigils (@@ and <<| |>>). Resource references (e.g., File["/etc/passwd"]) now have to be capitalized, in fitting in with capitalizing type operations. As usual, lots of other smaller fixes, but most of the work was in the language. 0.19.3 Fixing a bug in server/master.rb that causes the hostname not to be available in locally-executed manifests. 0.19.2 Fixing a few smaller bugs, notably in the reports system. Refreshed objects now generate an event, which can result in further refreshes of other objects. 0.19.1 Fixing two critical bugs: User management works again and cron jobs are no longer added to all user accounts. 0.19.0 Added provider support. Added support for %h, %H, and %d expansion in fileserver.conf. Added Certificate Revocation support. Made dynamic loading pervasive -- nearly every aspect of Puppet will now automatically load new instances (e.g., types, providers, and reports). Added support for automatic distribution of facts and plugins (custom types). 0.18.4 Another bug-fix release. The most import bug fixed is that cronjobs again work even with initially empty crontabs. 0.18.3 Mostly a bug-fix release; fixed small bugs in the functionality added in 0.18.2. 0.18.2 Added templating support. Added reporting. Added gem and blastwave packaging support. 0.18.1 Added signal handlers for HUP, so both client and server deal correctly with it. Added signal handler for USR1, which triggers a run on the client. As usual, fixed many bugs. Significant fixes to puppetrun -- it should behave much more correctly now. Added "fail" function which throws a syntax error if it's encountered. Added plugin downloading from the central server to the client. It must be enabled with --pluginsync. Added support for FreeBSD's special "@daily" cron schedules. Correctly handling spaces in file sources. Moved documentation into svn tree. 0.18.0 Added support for a "default" node. When multiple nodes are specified, they must now be comma-separated (this introduces a language incompatibility). Failed dependencies cause dependent objects within the same transaction not to run. Many updates to puppetrun Many bug fixes Function names are no longer reserved words. Links can now replace files. 0.17.2 Added "puppetrun" application and associated runner server and client classes. Fixed cron support so it better supports valid values and environment settings. 0.17.1 Fixing a bug requiring rails on all Debian boxes Fixing a couple of other small bugs 0.17.0 Adding ActiveRecord integration on the server Adding export/collect functionality Fixing many bugs 0.16.5 Fixing a critical bug in importing classes from other files Fixing nodename handling to actually allow dashes 0.16.4 Fixing a critical bug in puppetd when acquiring a certificate for the first time 0.16.3 Some significant bug fixes Modified puppetd so that it can now function as an agent independent of a puppetmasterd process, e.g., using the PuppetShow web application. 0.16.2 Modified some of the AST classes so that class names, definition names, and node names are all set within the code being evaluated, so 'tagged(name)' returns true while evaluating 'name', for instance. Added '--clean' argument to puppetca to remove all traces of a given client. 0.16.1 Added 'tagged' and 'defined' functions. Moved all functions to a general framework that makes it very easy to add new functions. 0.16.0 Added 'tag' keyword/function. Added FreeBSD Ports support Added 'pelement' server for sending or receiving Puppet objects, although none of the executables use it yet. 0.15.3 Fixed many bugs in :exec, including adding support for arrays of checks Added autoloading for types and service variants (e.g., you can now just create a new type in the appropriate location and use it in Puppet, without modifying the core Puppet libs). 0.15.2 Added darwinport, Apple .pkg, and freebsd package types Added 'mount type Host facts are now set at the top scope (Bug #103) Added -e (inline exection) flag to 'puppet' executable Many small bug fixes 0.15.1 Fixed 'yum' installs so that they successfully upgrade packages. Fixed puppetmasterd.conf file so group settings take. 0.15.0 Upped the minor release because the File server is incompatible with 0.14, because it now handles links. The 'symlink' type is deprecated (but still present), in favor of using files with the 'target' parameter. Unset variables no longer throw an error, they just return an empty string You can now specify tags to restrict which objects run during a given run. You can also specify to skip running against the cached copy when there's a failure, which is useful for testing new configurations. RPMs and Sun packages can now install, as long as they specify a package location, and they'll automatically upgrade if you point them to a new file with an upgrade. Multiple bug fixes. 0.14.1 Fixed a couple of small logging bugs Fixed a bug with handling group ownership of links 0.14.0 Added some ability to selectively manage symlinks when doing file management Many bug fixes Variables can now be used as the test values in case statements and selectors Bumping a minor release number because 0.13.4 introduced a protocol incompatibility and should have had a minor rev bump 0.13.6 Many, many small bug fixes FreeBSD user/group support has been added The configuration system has been rewritten so that daemons can now generate and repair the files and directories they need. (Fixed bug #68.) Fixed the element override issues; now only subclasses can override values. 0.13.5 Fixed packages so types can be specified Added 'enable' state to services, although it does not work everywhere yet 0.13.4 A few important bug fixes, mostly in the parser. 0.13.3 Changed transactions to be one-stage instead of two Changed all types to use self[:name] instead of self.name, to support the symbolic naming implemented in 0.13.1 0.13.2 Changed package[answerfile] to package[adminfile], and added package[responsefile] Fixed a bunch of internal functions to behave more consistently and usefully 0.13.1 Fixed RPM spec files to create puppet user and group (lutter) Fixed crontab reading and writing (luke) Added symbolic naming in the language (luke) 0.13.0 Added support for configuration files. Even more bug fixes, including the infamous 'frozen object' bug, which was a problem with 'waitforcert'. David Lutterkort got RPM into good shape. 0.12.0 Added Scheduling, and many bug fixes, of course. 0.11.2 Fixed bugs related to specifying arrays of requirements Fixed a key bug in retrieving checksums Fixed lots of usability bugs Added 'fail' methods that automatically add file and line info when possible, and converted many errors to use that method 0.11.1 Fixed bug with recursive copying with 'ignore' set. Added OpenBSD package support. 0.11.0 Added 'ensure' state to many elements. Modified puppetdoc to correctly handle indentation and such. Significantly rewrote much of the builtin documentation to take advantage of the new features in puppetdoc, including many examples. 0.10.2 Added SMF support Added autorequire functionality, with specific support for exec and file Exec elements autorequire any mentioned files, including the scripts, along with their CWDs. Files autorequire any parent directories. Added 'alias' metaparam. Fixed dependencies so they don't depend on file order. 0.10.1 Added Solaris package support and changed puppetmasterd to run as a non-root user. 0.10.0 Significant refactoring of how types, states, and parameters work, including breaking out parameters into a separate class. This refactoring did not introduce much new functionality, but made extension of Puppet significantly easier Also, fixed the bug with 'waitforcert' in puppetd. 0.9.4 Small fix to wrap the StatusServer class in the checks for required classes. 0.9.3 Fixed some significant bugs in cron job management. 0.9.2 Second Public Beta 0.9.0 First Public Beta diff --git a/lib/puppet/parser/compile.rb b/lib/puppet/parser/compile.rb index fdd0cbcf2..c065e3f38 100644 --- a/lib/puppet/parser/compile.rb +++ b/lib/puppet/parser/compile.rb @@ -1,509 +1,512 @@ # Created by Luke A. Kanies on 2007-08-13. # Copyright (c) 2007. All rights reserved. require 'puppet/external/gratr/digraph' require 'puppet/external/gratr/import' require 'puppet/external/gratr/dot' require 'puppet/node' require 'puppet/node/catalog' require 'puppet/util/errors' # Maintain a graph of scopes, along with a bunch of data # about the individual catalog we're compiling. class Puppet::Parser::Compile include Puppet::Util include Puppet::Util::Errors attr_reader :parser, :node, :facts, :collections, :catalog, :node_scope # Add a collection to the global list. def add_collection(coll) @collections << coll end # Do we use nodes found in the code, vs. the external node sources? def ast_nodes? parser.nodes.length > 0 end # Store the fact that we've evaluated a class, and store a reference to # the scope in which it was evaluated, so that we can look it up later. def class_set(name, scope) if existing = @class_scopes[name] if existing.nodescope? or scope.nodescope? raise Puppet::ParseError, "Cannot have classes, nodes, or definitions with the same name" else raise Puppet::DevError, "Somehow evaluated the same class twice" end end @class_scopes[name] = scope @catalog.add_class(name) unless name == "" end # Return the scope associated with a class. This is just here so # that subclasses can set their parent scopes to be the scope of # their parent class, and it's also used when looking up qualified # variables. def class_scope(klass) # They might pass in either the class or class name if klass.respond_to?(:classname) @class_scopes[klass.classname] else @class_scopes[klass] end end # Return a list of all of the defined classes. def classlist return @catalog.classes end # Compile our catalog. This mostly revolves around finding and evaluating classes. # This is the main entry into our catalog. def compile # Set the client's parameters into the top scope. set_node_parameters() evaluate_main() evaluate_ast_node() evaluate_node_classes() evaluate_generators() fail_on_unevaluated() finish() if Puppet[:storeconfigs] store() end return @catalog end # LAK:FIXME There are no tests for this. def delete_collection(coll) @collections.delete(coll) if @collections.include?(coll) end # LAK:FIXME There are no tests for this. def delete_resource(resource) @resource_table.delete(resource.ref) if @resource_table.include?(resource.ref) end # Return the node's environment. def environment unless defined? @environment if node.environment and node.environment != "" @environment = node.environment else @environment = nil end end @environment end # Evaluate all of the classes specified by the node. def evaluate_node_classes evaluate_classes(@node.classes, topscope) end # Evaluate each specified class in turn. If there are any classes we can't # find, just tag the catalog and move on. This method really just # creates resource objects that point back to the classes, and then the # resources are themselves evaluated later in the process. def evaluate_classes(classes, scope, lazy_evaluate = true) unless scope.source raise Puppet::DevError, "No source for scope passed to evaluate_classes" end found = [] classes.each do |name| # If we can find the class, then make a resource that will evaluate it. if klass = scope.findclass(name) + found << name and next if class_scope(klass) + # Create a resource to model this class, and then add it to the list # of resources. resource = Puppet::Parser::Resource.new(:type => "class", :title => klass.classname, :scope => scope, :source => scope.source) + store_resource(scope, resource) # If they've disabled lazy evaluation (which the :include function does), # then evaluate our resource immediately. resource.evaluate unless lazy_evaluate @catalog.tag(klass.classname) found << name else Puppet.info "Could not find class %s for %s" % [name, node.name] @catalog.tag(name) end end found end # Return a resource by either its ref or its type and title. def findresource(string, name = nil) string = "%s[%s]" % [string.capitalize, name] if name @resource_table[string] end # Set up our compile. We require a parser # and a node object; the parser is so we can look up classes # and AST nodes, and the node has all of the client's info, # like facts and environment. def initialize(node, parser, options = {}) @node = node @parser = parser options.each do |param, value| begin send(param.to_s + "=", value) rescue NoMethodError raise ArgumentError, "Compile objects do not accept %s" % param end end initvars() init_main() end # Create a new scope, with either a specified parent scope or # using the top scope. Adds an edge between the scope and # its parent to the graph. def newscope(parent, options = {}) parent ||= topscope options[:compile] = self options[:parser] ||= self.parser scope = Puppet::Parser::Scope.new(options) @scope_graph.add_edge!(parent, scope) scope end # Find the parent of a given scope. Assumes scopes only ever have # one in edge, which will always be true. def parent(scope) if ary = @scope_graph.adjacent(scope, :direction => :in) and ary.length > 0 ary[0] else nil end end # Return any overrides for the given resource. def resource_overrides(resource) @resource_overrides[resource.ref] end # Return a list of all resources. def resources @resource_table.values end # Store a resource override. def store_override(override) override.override = true # If possible, merge the override in immediately. if resource = @resource_table[override.ref] resource.merge(override) else # Otherwise, store the override for later; these # get evaluated in Resource#finish. @resource_overrides[override.ref] << override end end # Store a resource in our resource table. def store_resource(scope, resource) # This might throw an exception verify_uniqueness(resource) # Store it in the global table. @resource_table[resource.ref] = resource # And in the resource graph. At some point, this might supercede # the global resource table, but the table is a lot faster # so it makes sense to maintain for now. @catalog.add_edge!(scope.resource, resource) end # The top scope is usually the top-level scope, but if we're using AST nodes, # then it is instead the node's scope. def topscope node_scope || @topscope end private # If ast nodes are enabled, then see if we can find and evaluate one. def evaluate_ast_node return unless ast_nodes? # Now see if we can find the node. astnode = nil @node.names.each do |name| break if astnode = @parser.nodes[name.to_s.downcase] end unless (astnode ||= @parser.nodes["default"]) raise Puppet::ParseError, "Could not find default node or by name with '%s'" % node.names.join(", ") end # Create a resource to model this node, and then add it to the list # of resources. resource = Puppet::Parser::Resource.new(:type => "node", :title => astnode.classname, :scope => topscope, :source => topscope.source) store_resource(topscope, resource) @catalog.tag(astnode.classname) resource.evaluate # Now set the node scope appropriately, so that :topscope can # behave differently. @node_scope = class_scope(astnode) end # Evaluate our collections and return true if anything returned an object. # The 'true' is used to continue a loop, so it's important. def evaluate_collections return false if @collections.empty? found_something = false exceptwrap do # We have to iterate over a dup of the array because # collections can delete themselves from the list, which # changes its length and causes some collections to get missed. @collections.dup.each do |collection| found_something = true if collection.evaluate end end return found_something end # Make sure all of our resources have been evaluated into native resources. # We return true if any resources have, so that we know to continue the # evaluate_generators loop. def evaluate_definitions exceptwrap do if ary = unevaluated_resources ary.each do |resource| resource.evaluate end # If we evaluated, let the loop know. return true else return false end end end # Iterate over collections and resources until we're sure that the whole # compile is evaluated. This is necessary because both collections # and defined resources can generate new resources, which themselves could # be defined resources. def evaluate_generators count = 0 loop do done = true # Call collections first, then definitions. done = false if evaluate_collections done = false if evaluate_definitions break if done if count > 1000 raise Puppet::ParseError, "Somehow looped more than 1000 times while evaluating host catalog" end end end # Find and evaluate our main object, if possible. def evaluate_main @main = @parser.findclass("", "") || @parser.newclass("") @topscope.source = @main @main_resource = Puppet::Parser::Resource.new(:type => "class", :title => :main, :scope => @topscope, :source => @main) @topscope.resource = @main_resource @catalog.add_vertex!(@main_resource) @resource_table["Class[main]"] = @main_resource @main_resource.evaluate end # Make sure the entire catalog is evaluated. def fail_on_unevaluated fail_on_unevaluated_overrides fail_on_unevaluated_resource_collections end # If there are any resource overrides remaining, then we could # not find the resource they were supposed to override, so we # want to throw an exception. def fail_on_unevaluated_overrides remaining = [] @resource_overrides.each do |name, overrides| remaining += overrides end unless remaining.empty? fail Puppet::ParseError, "Could not find object(s) %s" % remaining.collect { |o| o.ref }.join(", ") end end # Make sure we don't have any remaining collections that specifically # look for resources, because we want to consider those to be # parse errors. def fail_on_unevaluated_resource_collections remaining = [] @collections.each do |coll| # We're only interested in the 'resource' collections, # which result from direct calls of 'realize'. Anything # else is allowed not to return resources. # Collect all of them, so we have a useful error. if r = coll.resources if r.is_a?(Array) remaining += r else remaining << r end end end unless remaining.empty? raise Puppet::ParseError, "Failed to realize virtual resources %s" % remaining.join(', ') end end # Make sure all of our resources and such have done any last work # necessary. def finish @resource_table.each { |name, resource| resource.finish if resource.respond_to?(:finish) } end # Initialize the top-level scope, class, and resource. def init_main # Create our initial scope and a resource that will evaluate main. @topscope = Puppet::Parser::Scope.new(:compile => self, :parser => self.parser) @scope_graph.add_vertex!(@topscope) end # Set up all of our internal variables. def initvars # The table for storing class singletons. This will only actually # be used by top scopes and node scopes. @class_scopes = {} # The table for all defined resources. @resource_table = {} # The list of objects that will available for export. @exported_resources = {} # The list of overrides. This is used to cache overrides on objects # that don't exist yet. We store an array of each override. @resource_overrides = Hash.new do |overs, ref| overs[ref] = [] end # The list of collections that have been created. This is a global list, # but they each refer back to the scope that created them. @collections = [] # A list of tags we've generated; most class names. @tags = [] # A graph for maintaining scope relationships. @scope_graph = GRATR::Digraph.new # For maintaining the relationship between scopes and their resources. @catalog = Puppet::Node::Catalog.new(@node.name) @catalog.version = @parser.version end # Set the node's parameters into the top-scope as variables. def set_node_parameters node.parameters.each do |param, value| @topscope.setvar(param, value) end end # Store the catalog into the database. def store unless Puppet.features.rails? raise Puppet::Error, "storeconfigs is enabled but rails is unavailable" end unless ActiveRecord::Base.connected? Puppet::Rails.connect end # We used to have hooks here for forking and saving, but I don't # think it's worth retaining at this point. store_to_active_record(@node, @resource_table.values) end # Do the actual storage. def store_to_active_record(node, resources) begin # We store all of the objects, even the collectable ones benchmark(:info, "Stored catalog for #{node.name}") do Puppet::Rails::Host.transaction do Puppet::Rails::Host.store(node, resources) end end rescue => detail if Puppet[:trace] puts detail.backtrace end Puppet.err "Could not store configs: %s" % detail.to_s end end # Return an array of all of the unevaluated resources. These will be definitions, # which need to get evaluated into native resources. def unevaluated_resources ary = @resource_table.find_all do |name, object| ! object.builtin? and ! object.evaluated? end.collect { |name, object| object } if ary.empty? return nil else return ary end end # Verify that the given resource isn't defined elsewhere. def verify_uniqueness(resource) # Short-curcuit the common case, unless existing_resource = @resource_table[resource.ref] return true end if typeclass = Puppet::Type.type(resource.type) and ! typeclass.isomorphic? Puppet.info "Allowing duplicate %s" % typeclass.name return true end # Either it's a defined type, which are never # isomorphic, or it's a non-isomorphic type, so # we should throw an exception. msg = "Duplicate definition: %s is already defined" % resource.ref if existing_resource.file and existing_resource.line msg << " in file %s at line %s" % [existing_resource.file, existing_resource.line] end if resource.line or resource.file msg << "; cannot redefine" end raise Puppet::ParseError.new(msg) end end diff --git a/spec/unit/parser/compile.rb b/spec/unit/parser/compile.rb index 2ae99b5fd..092bece0c 100755 --- a/spec/unit/parser/compile.rb +++ b/spec/unit/parser/compile.rb @@ -1,268 +1,281 @@ #!/usr/bin/env ruby require File.dirname(__FILE__) + '/../../spec_helper' describe Puppet::Parser::Compile, " when compiling" do before do @node = stub 'node', :name => 'mynode' @parser = stub 'parser', :version => "1.0" @compile = Puppet::Parser::Compile.new(@node, @parser) end def compile_methods [:set_node_parameters, :evaluate_main, :evaluate_ast_node, :evaluate_node_classes, :evaluate_generators, :fail_on_unevaluated, :finish, :store, :extract] end # Stub all of the main compile methods except the ones we're specifically interested in. def compile_stub(*except) (compile_methods - except).each { |m| @compile.stubs(m) } end it "should set node parameters as variables in the top scope" do params = {"a" => "b", "c" => "d"} @node.stubs(:parameters).returns(params) compile_stub(:set_node_parameters) @compile.compile @compile.topscope.lookupvar("a").should == "b" @compile.topscope.lookupvar("c").should == "d" end it "should evaluate any existing classes named in the node" do classes = %w{one two three four} main = stub 'main' one = stub 'one', :classname => "one" three = stub 'three', :classname => "three" @node.stubs(:name).returns("whatever") @node.stubs(:classes).returns(classes) @compile.expects(:evaluate_classes).with(classes, @compile.topscope) @compile.send :evaluate_node_classes end it "should enable ast_nodes if the parser has any nodes" do @parser.expects(:nodes).returns(:one => :yay) @compile.ast_nodes?.should be_true end it "should disable ast_nodes if the parser has no nodes" do @parser.expects(:nodes).returns({}) @compile.ast_nodes?.should be_false end end describe Puppet::Parser::Compile, " when evaluating classes" do before do @node = stub 'node', :name => 'mynode' @parser = stub 'parser', :version => "1.0" @scope = stub 'scope', :source => mock("source") @compile = Puppet::Parser::Compile.new(@node, @parser) end it "should fail if there's no source listed for the scope" do scope = stub 'scope', :source => nil proc { @compile.evaluate_classes(%w{one two}, scope) }.should raise_error(Puppet::DevError) end it "should tag the catalog with the name of each not-found class" do @compile.catalog.expects(:tag).with("notfound") @scope.expects(:findclass).with("notfound").returns(nil) @compile.evaluate_classes(%w{notfound}, @scope) end end describe Puppet::Parser::Compile, " when evaluating collections" do before do @node = stub 'node', :name => 'mynode' @parser = stub 'parser', :version => "1.0" @scope = stub 'scope', :source => mock("source") @compile = Puppet::Parser::Compile.new(@node, @parser) end it "should evaluate each collection" do 2.times { |i| coll = mock 'coll%s' % i @compile.add_collection(coll) # This is the hard part -- we have to emulate the fact that # collections delete themselves if they are done evaluating. coll.expects(:evaluate).with do @compile.delete_collection(coll) end } @compile.class.publicize_methods(:evaluate_collections) { @compile.evaluate_collections } end end describe Puppet::Parser::Compile, " when evaluating found classes" do before do @node = stub 'node', :name => 'mynode' @parser = stub 'parser', :version => "1.0" @scope = stub 'scope', :source => mock("source") @compile = Puppet::Parser::Compile.new(@node, @parser) @class = stub 'class', :classname => "my::class" @scope.stubs(:findclass).with("myclass").returns(@class) - @resource = mock 'resource' + @resource = stub 'resource', :ref => 'Class[myclass]' end it "should create a resource for each found class" do @compile.catalog.stubs(:tag) @compile.stubs :store_resource Puppet::Parser::Resource.expects(:new).with(:scope => @scope, :source => @scope.source, :title => "my::class", :type => "class").returns(@resource) @compile.evaluate_classes(%w{myclass}, @scope) end it "should store each created resource in the compile" do @compile.catalog.stubs(:tag) @compile.expects(:store_resource).with(@scope, @resource) Puppet::Parser::Resource.stubs(:new).returns(@resource) @compile.evaluate_classes(%w{myclass}, @scope) end it "should tag the catalog with the fully-qualified name of each found class" do @compile.catalog.expects(:tag).with("my::class") @compile.stubs(:store_resource) Puppet::Parser::Resource.stubs(:new).returns(@resource) @compile.evaluate_classes(%w{myclass}, @scope) end it "should not evaluate the resources created for found classes unless asked" do @compile.catalog.stubs(:tag) @compile.stubs(:store_resource) @resource.expects(:evaluate).never Puppet::Parser::Resource.stubs(:new).returns(@resource) @compile.evaluate_classes(%w{myclass}, @scope) end it "should immediately evaluate the resources created for found classes when asked" do @compile.catalog.stubs(:tag) @compile.stubs(:store_resource) @resource.expects(:evaluate) Puppet::Parser::Resource.stubs(:new).returns(@resource) @compile.evaluate_classes(%w{myclass}, @scope, false) end + it "should skip classes that have already been evaluated" do + @compile.catalog.stubs(:tag) + + @compile.expects(:class_scope).with(@class).returns("something") + + @compile.expects(:store_resource).never + + @resource.expects(:evaluate).never + + Puppet::Parser::Resource.expects(:new).never + @compile.evaluate_classes(%w{myclass}, @scope, false) + end + it "should return the list of found classes" do @compile.catalog.stubs(:tag) @compile.stubs(:store_resource) @scope.stubs(:findclass).with("notfound").returns(nil) Puppet::Parser::Resource.stubs(:new).returns(@resource) @compile.evaluate_classes(%w{myclass notfound}, @scope).should == %w{myclass} end end describe Puppet::Parser::Compile, " when evaluating AST nodes with no AST nodes present" do before do @node = stub 'node', :name => "foo" @parser = stub 'parser', :version => "1.0", :nodes => {} @compile = Puppet::Parser::Compile.new(@node, @parser) end it "should do nothing" do @compile.expects(:ast_nodes?).returns(false) @compile.parser.expects(:nodes).never Puppet::Parser::Resource.expects(:new).never @compile.send(:evaluate_ast_node) end end describe Puppet::Parser::Compile, " when evaluating AST nodes with AST nodes present" do before do @node = stub 'node', :name => "foo" @parser = stub 'parser', :version => "1.0", :nodes => {} @compile = Puppet::Parser::Compile.new(@node, @parser) @nodes = mock 'node_hash' @compile.stubs(:ast_nodes?).returns(true) @compile.parser.stubs(:nodes).returns(@nodes) # Set some names for our test @node.stubs(:names).returns(%w{a b c}) @nodes.stubs(:[]).with("a").returns(nil) @nodes.stubs(:[]).with("b").returns(nil) @nodes.stubs(:[]).with("c").returns(nil) # It should check this last, of course. @nodes.stubs(:[]).with("default").returns(nil) end it "should fail if the named node cannot be found" do proc { @compile.send(:evaluate_ast_node) }.should raise_error(Puppet::ParseError) end it "should create a resource for the first node class matching the node name" do node_class = stub 'node', :classname => "c" @nodes.stubs(:[]).with("c").returns(node_class) node_resource = stub 'node resource', :ref => "Node[c]", :evaluate => nil Puppet::Parser::Resource.expects(:new).with { |args| args[:title] == "c" and args[:type] == "node" }.returns(node_resource) @compile.send(:evaluate_ast_node) end it "should match the default node if no matching node can be found" do node_class = stub 'node', :classname => "default" @nodes.stubs(:[]).with("default").returns(node_class) node_resource = stub 'node resource', :ref => "Node[default]", :evaluate => nil Puppet::Parser::Resource.expects(:new).with { |args| args[:title] == "default" and args[:type] == "node" }.returns(node_resource) @compile.send(:evaluate_ast_node) end it "should tag the catalog with the found node name" do node_class = stub 'node', :classname => "c" @nodes.stubs(:[]).with("c").returns(node_class) node_resource = stub 'node resource', :ref => "Node[c]", :evaluate => nil Puppet::Parser::Resource.stubs(:new).returns(node_resource) @compile.catalog.expects(:tag).with("c") @compile.send(:evaluate_ast_node) end it "should evaluate the node resource immediately rather than using lazy evaluation" do node_class = stub 'node', :classname => "c" @nodes.stubs(:[]).with("c").returns(node_class) node_resource = stub 'node resource', :ref => "Node[c]" Puppet::Parser::Resource.stubs(:new).returns(node_resource) node_resource.expects(:evaluate) @compile.send(:evaluate_ast_node) end it "should set the node's scope as the top scope" do node_class = stub 'node', :classname => "c" @nodes.stubs(:[]).with("c").returns(node_class) node_resource = stub 'node resource', :ref => "Node[c]" Puppet::Parser::Resource.stubs(:new).returns(node_resource) # The #evaluate method normally does this. @compile.class_set(node_class.classname, :my_node_scope) node_resource.stubs(:evaluate) @compile.send(:evaluate_ast_node) @compile.topscope.should == :my_node_scope end end