Class | MCollective::RPC::Client |
In: |
lib/mcollective/rpc/client.rb
|
Parent: | Object |
The main component of the Simple RPC client system, this wraps around MCollective::Client and just brings in a lot of convention and standard approached.
agent | [R] | |
client | [R] | |
config | [RW] | |
ddl | [R] | |
discovery_timeout | [RW] | |
filter | [RW] | |
limit_targets | [R] | |
output_format | [R] | |
progress | [RW] | |
stats | [R] | |
timeout | [RW] | |
verbose | [RW] |
Creates a stub for a remote agent, you can pass in an options array in the flags which will then be used else it will just create a default options array with filtering enabled based on the standard command line use.
rpc = RPC::Client.new("rpctest", :configfile => "client.cfg", :options => options)
You typically would not call this directly you‘d use MCollective::RPC#rpcclient instead which is a wrapper around this that can be used as a Mixin
# File lib/mcollective/rpc/client.rb, line 19 19: def initialize(agent, flags = {}) 20: if flags.include?(:options) 21: options = flags[:options] 22: 23: elsif @@initial_options 24: options = Marshal.load(@@initial_options) 25: 26: else 27: oparser = MCollective::Optionparser.new({:verbose => false, :progress_bar => true, :mcollective_limit_targets => false}, "filter") 28: 29: options = oparser.parse do |parser, options| 30: if block_given? 31: yield(parser, options) 32: end 33: 34: Helpers.add_simplerpc_options(parser, options) 35: end 36: 37: @@initial_options = Marshal.dump(options) 38: end 39: 40: @stats = Stats.new 41: @agent = agent 42: @discovery_timeout = options[:disctimeout] 43: @timeout = options[:timeout] 44: @verbose = options[:verbose] 45: @filter = options[:filter] 46: @config = options[:config] 47: @discovered_agents = nil 48: @progress = options[:progress_bar] 49: @limit_targets = options[:mcollective_limit_targets] 50: @output_format = options[:output_format] || :console 51: @force_direct_request = false 52: 53: agent_filter agent 54: 55: @client = MCollective::Client.new(@config) 56: @client.options = options 57: 58: @collective = @client.collective 59: 60: # if we can find a DDL for the service override 61: # the timeout of the client so we always magically 62: # wait appropriate amounts of time. 63: # 64: # We add the discovery timeout to the ddl supplied 65: # timeout as the discovery timeout tends to be tuned 66: # for local network conditions and fact source speed 67: # which would other wise not be accounted for and 68: # some results might get missed. 69: # 70: # We do this only if the timeout is the default 5 71: # seconds, so that users cli overrides will still 72: # get applied 73: begin 74: @ddl = DDL.new(agent) 75: @timeout = @ddl.meta[:timeout] + @discovery_timeout if @timeout == 5 76: rescue Exception => e 77: Log.debug("Could not find DDL: #{e}") 78: @ddl = nil 79: end 80: 81: STDERR.sync = true 82: STDOUT.sync = true 83: end
Sets the agent filter
# File lib/mcollective/rpc/client.rb, line 302 302: def agent_filter(agent) 303: @filter["agent"] << agent 304: @filter["agent"].compact! 305: reset 306: end
Sets the class filter
# File lib/mcollective/rpc/client.rb, line 278 278: def class_filter(klass) 279: @filter["cf_class"] << klass 280: @filter["cf_class"].compact! 281: reset 282: end
Sets the collective we are communicating with
# File lib/mcollective/rpc/client.rb, line 421 421: def collective=(c) 422: @collective = c 423: @client.options[:collective] = c 424: end
Set a compound filter
# File lib/mcollective/rpc/client.rb, line 316 316: def compound_filter(filter) 317: @filter["compound"] = Matcher::Parser.new(filter).execution_stack 318: reset 319: end
Constructs custom requests with custom filters and discovery data the idea is that this would be used in web applications where you might be using a cached copy of data provided by a registration agent to figure out on your own what nodes will be responding and what your filter would be.
This will help you essentially short circuit the traditional cycle of:
mc discover / call / wait for discovered nodes
by doing discovery however you like, contructing a filter and a list of nodes you expect responses from.
Other than that it will work exactly like a normal call, blocks will behave the same way, stats will be handled the same way etcetc
If you just wanted to contact one machine for example with a client that already has other filter options setup you can do:
puppet.custom_request("runonce", {}, ["your.box.com"], {:identity => "your.box.com"})
This will do runonce action on just ‘your.box.com’, no discovery will be done and after receiving just one response it will stop waiting for responses
If direct_addressing is enabled in the config file you can provide an empty hash as a filter, this will force that request to be a directly addressed request which technically does not need filters. If you try to use this mode with direct addressing disabled an exception will be raise
# File lib/mcollective/rpc/client.rb, line 232 232: def custom_request(action, args, expected_agents, filter = {}, &block) 233: @ddl.validate_request(action, args) if @ddl 234: 235: if filter == {} && !Config.instance.direct_addressing 236: raise "Attempted to do a filterless custom_request without direct_addressing enabled, preventing unexpected call to all nodes" 237: end 238: 239: 240: @stats.reset 241: 242: custom_filter = Util.empty_filter 243: custom_options = options.clone 244: 245: # merge the supplied filter with the standard empty one 246: # we could just use the merge method but I want to be sure 247: # we dont merge in stuff that isnt actually valid 248: ["identity", "fact", "agent", "cf_class", "compound"].each do |ftype| 249: if filter.include?(ftype) 250: custom_filter[ftype] = [filter[ftype], custom_filter[ftype]].flatten 251: end 252: end 253: 254: # ensure that all filters at least restrict the call to the agent we're a proxy for 255: custom_filter["agent"] << @agent unless custom_filter["agent"].include?(@agent) 256: custom_options[:filter] = custom_filter 257: 258: # Fake out the stats discovery would have put there 259: @stats.discovered_agents([expected_agents].flatten) 260: 261: # Handle fire and forget requests 262: if args.include?(:process_results) && args[:process_results] == false 263: return fire_and_forget_request(action, args, custom_filter) 264: end 265: 266: # Now do a call pretty much exactly like in method_missing except with our own 267: # options and discovery magic 268: if block_given? 269: call_agent(action, args, custom_options, [expected_agents].flatten) do |r| 270: block.call(r) 271: end 272: else 273: call_agent(action, args, custom_options, [expected_agents].flatten) 274: end 275: end
Disconnects cleanly from the middleware
# File lib/mcollective/rpc/client.rb, line 86 86: def disconnect 87: @client.disconnect 88: end
Does discovery based on the filters set, if a discovery was previously done return that else do a new discovery.
Alternatively if identity filters are given and none of them are regular expressions then just use the provided data as discovered data, avoiding discovery
Discovery can be forece if direct_addressing is enabled by passing in an array of hosts with :hosts or JSON data like those produced by mcollective rpc JSON output
Will show a message indicating its doing discovery if running verbose or if the :verbose flag is passed in.
Use reset to force a new discovery
# File lib/mcollective/rpc/client.rb, line 348 348: def discover(flags={}) 349: flags.include?(:verbose) ? verbose = flags[:verbose] : verbose = @verbose 350: 351: verbose = false unless @output_format == :console 352: 353: unless @discovered_agents 354: # if either hosts or json is supplied try to figure out discovery data from there 355: # if direct_addressing is not enabled this is a critical error as the user might 356: # not have supplied filters so raise an exception 357: if flags[:hosts] || flags[:json] 358: raise "Can only supply discovery data if direct_addressing is enabled" unless Config.instance.direct_addressing 359: 360: hosts = [] 361: 362: if flags[:hosts] 363: hosts = Helpers.extract_hosts_from_array(flags[:hosts]) 364: elsif flags[:json] 365: hosts = Helpers.extract_hosts_from_json(flags[:json]) 366: end 367: 368: raise "Could not find any hosts in discovery data provided" if hosts.empty? 369: 370: @discovered_agents = hosts 371: @force_direct_request = true 372: 373: # if an identity filter is supplied and it is all strings no regex we can use that 374: # as discovery data, technically the identity filter is then redundant if we are 375: # in direct addressing mode and we could empty it out but this use case should 376: # only really be for a few -I's on the cli 377: # 378: # For safety we leave the filter in place for now, that way we can support this 379: # enhancement also in broadcast mode 380: elsif options[:filter]["identity"].size > 0 381: regex_filters = options[:filter]["identity"].select{|i| i.match("^\/")}.size 382: 383: if regex_filters == 0 384: @discovered_agents = options[:filter]["identity"].clone 385: @force_direct_request = true if Config.instance.direct_addressing 386: end 387: end 388: end 389: 390: # All else fails we do it the hard way using a traditional broadcast 391: unless @discovered_agents 392: @stats.time_discovery :start 393: 394: STDERR.print("Determining the amount of hosts matching filter for #{discovery_timeout} seconds .... ") if verbose 395: @discovered_agents = @client.discover(@filter, @discovery_timeout) 396: @force_direct_request = false 397: STDERR.puts(@discovered_agents.size) if verbose 398: 399: @stats.time_discovery :end 400: end 401: 402: @stats.discovered_agents(@discovered_agents) 403: RPC.discovered(@discovered_agents) 404: 405: @discovered_agents 406: end
Sets the fact filter
# File lib/mcollective/rpc/client.rb, line 285 285: def fact_filter(fact, value=nil, operator="=") 286: return if fact.nil? 287: return if fact == false 288: 289: if value.nil? 290: parsed = Util.parse_fact_string(fact) 291: @filter["fact"] << parsed unless parsed == false 292: else 293: parsed = Util.parse_fact_string("#{fact}#{operator}#{value}") 294: @filter["fact"] << parsed unless parsed == false 295: end 296: 297: @filter["fact"].compact! 298: reset 299: end
Sets the identity filter
# File lib/mcollective/rpc/client.rb, line 309 309: def identity_filter(identity) 310: @filter["identity"] << identity 311: @filter["identity"].compact! 312: reset 313: end
Sets and sanity checks the limit_targets variable used to restrict how many nodes we‘ll target
# File lib/mcollective/rpc/client.rb, line 428 428: def limit_targets=(limit) 429: if limit.is_a?(String) 430: raise "Invalid limit specified: #{limit} valid limits are /^\d+%*$/" unless limit =~ /^\d+%*$/ 431: @limit_targets = limit 432: elsif limit.respond_to?(:to_i) 433: limit = limit.to_i 434: limit = 1 if limit == 0 435: @limit_targets = limit 436: else 437: raise "Don't know how to handle limit of type #{limit.class}" 438: end 439: end
Magic handler to invoke remote methods
Once the stub is created using the constructor or the RPC#rpcclient helper you can call remote actions easily:
ret = rpc.echo(:msg => "hello world")
This will call the ‘echo’ action of the ‘rpctest’ agent and return the result as an array, the array will be a simplified result set from the usual full MCollective::Client#req with additional error codes and error text:
{
:sender => "remote.box.com", :statuscode => 0, :statusmsg => "OK", :data => "hello world"
}
If :statuscode is 0 then everything went find, if it‘s 1 then you supplied the correct arguments etc but the request could not be completed, you‘ll find a human parsable reason in :statusmsg then.
Codes 2 to 5 maps directly to UnknownRPCAction, MissingRPCData, InvalidRPCData and UnknownRPCError see below for a description of those, in each case :statusmsg would be the reason for failure.
To get access to the full result of the MCollective::Client#req calls you can pass in a block:
rpc.echo(:msg => "hello world") do |resp| pp resp end
In this case resp will the result from MCollective::Client#req. Instead of returning simple text and codes as above you‘ll also need to handle the following exceptions:
UnknownRPCAction - There is no matching action on the agent MissingRPCData - You did not supply all the needed parameters for the action InvalidRPCData - The data you did supply did not pass validation UnknownRPCError - Some other error prevented the agent from running
During calls a progress indicator will be shown of how many results we‘ve received against how many nodes were discovered, you can disable this by setting progress to false:
rpc.progress = false
This supports a 2nd mode where it will send the SimpleRPC request and never handle the responses. It‘s a bit like UDP, it sends the request with the filter attached and you only get back the requestid, you have no indication about results.
You can invoke this using:
puts rpc.echo(:process_results => false)
This will output just the request id.
# File lib/mcollective/rpc/client.rb, line 179 179: def method_missing(method_name, *args, &block) 180: # set args to an empty hash if nothings given 181: args = args[0] 182: args = {} if args.nil? 183: 184: action = method_name.to_s 185: 186: @stats.reset 187: 188: @ddl.validate_request(action, args) if @ddl 189: 190: # Handle single target requests by doing discovery and picking 191: # a random node. Then do a custom request specifying a filter 192: # that will only match the one node. 193: if @limit_targets 194: target_nodes = pick_nodes_from_discovered(@limit_targets) 195: Log.debug("Picked #{target_nodes.join(',')} as limited target(s)") 196: 197: custom_request(action, args, target_nodes, {"identity" => /^(#{target_nodes.join('|')})$/}, &block) 198: else 199: # Normal agent requests as per client.action(args) 200: call_agent(action, args, options, &block) 201: end 202: end
Creates a suitable request hash for the SimpleRPC agent.
You‘d use this if you ever wanted to take care of sending requests on your own - perhaps via Client#sendreq if you didn‘t care for responses.
In that case you can just do:
msg = your_rpc.new_request("some_action", :foo => :bar) filter = your_rpc.filter your_rpc.client.sendreq(msg, msg[:agent], filter)
This will send a SimpleRPC request to the action some_action with arguments :foo = :bar, it will return immediately and you will have no indication at all if the request was receieved or not
Clearly the use of this technique should be limited and done only if your code requires such a thing
# File lib/mcollective/rpc/client.rb, line 118 118: def new_request(action, data) 119: callerid = PluginManager["security_plugin"].callerid 120: 121: {:agent => @agent, 122: :action => action, 123: :caller => callerid, 124: :data => data} 125: end
Provides a normal options hash like you would get from Optionparser
# File lib/mcollective/rpc/client.rb, line 410 410: def options 411: {:disctimeout => @discovery_timeout, 412: :timeout => @timeout, 413: :verbose => @verbose, 414: :filter => @filter, 415: :collective => @collective, 416: :output_format => @output_format, 417: :config => @config} 418: end
Resets various internal parts of the class, most importantly it clears out the cached discovery
# File lib/mcollective/rpc/client.rb, line 323 323: def reset 324: @discovered_agents = nil 325: end