The mod_aggregate system was added in the 2014.7.0 release of Salt and allows for runtime modification of the executing state data. Simply put, it allows for the data used by Salt's state system to be changed on the fly at runtime, kind of like a configuration management JIT compiler or a runtime import system. All in all, it makes Salt much more dynamic.
The best example is the
pkg state. One of the major requests in Salt has long
been adding the ability to install all packages defined at the same time. The
mod_aggregate system makes this a reality. While executing Salt's state system,
pkg state is reached the
mod_aggregate function in the state module
is called. For
pkg this function scans all of the other states that are slated
to run, and picks up the references to
pkgs, then adds them to
pkgs in the first state. The result is a single call to yum, apt-get,
pacman, etc as part of the first package install.
Since this option changes the basic behavior of the state runtime, after it is enabled states should be executed using test=True to ensure that the desired behavior is preserved.
The first way to enable aggregation is with a configuration option in either
the master or minion configuration files. Salt will invoke
the first time it encounters a state module that has aggregate support.
If this option is set in the master config it will apply to all state runs on all minions, if set in the minion config it will only apply to said minion.
Enable for all states:
Enable for only specific state modules:
state_aggregate: - pkg
The second way to enable aggregation is with the state-level
keyword. In this configuration, Salt will invoke the
the first time it encounters this keyword. Any additional occurrences of the
keyword will be ignored as the aggregation has already taken place.
The following example will trigger
mod_aggregate when the
state is processed resulting in a single call to the underlying package
lamp_stack: pkg.installed: - pkgs: - php - mysql-client - aggregate: True memcached: pkg.installed: - name: memcached
Adding a mod_aggregate routine to an existing state module only requires adding an additional function to the state module called mod_aggregate.
The mod_aggregate function just needs to accept three parameters and return the low data to use. Since mod_aggregate is working on the state runtime level it does need to manipulate low data.
The three parameters are low, chunks, and running. The low option is the low data for the state execution which is about to be called. The chunks is the list of all of the low data dictionaries which are being executed by the runtime and the running dictionary is the return data from all of the state executions which have already be executed.
This example, simplified from the pkg state, shows how to create mod_aggregate functions:
def mod_aggregate(low, chunks, running): ''' The mod_aggregate function which looks up all packages in the available low chunks and merges them into a single pkgs ref in the present low data ''' pkgs =  # What functions should we aggregate? agg_enabled = [ 'installed', 'latest', 'removed', 'purged', ] # The `low` data is just a dict with the state, function (fun) and # arguments passed in from the sls if low.get('fun') not in agg_enabled: return low # Now look into what other things are set to execute for chunk in chunks: # The state runtime uses "tags" to track completed jobs, it may # look familiar with the _|- tag = salt.utils.gen_state_tag(chunk) if tag in running: # Already ran the pkg state, skip aggregation continue if chunk.get('state') == 'pkg': if '__agg__' in chunk: continue # Check for the same function if chunk.get('fun') != low.get('fun'): continue # Pull out the pkg names! if 'pkgs' in chunk: pkgs.extend(chunk['pkgs']) chunk['__agg__'] = True elif 'name' in chunk: pkgs.append(chunk['name']) chunk['__agg__'] = True if pkgs: if 'pkgs' in low: low['pkgs'].extend(pkgs) else: low['pkgs'] = pkgs # The low has been modified and needs to be returned to the state # runtime for execution return low