Table of Contents
Eventor provides programmer with interface to create events, steps and associations of these artifacts with to create a flow.
It would be easier to show an example.
import eventor as evr import logging from acrilib import LoggerAddHostFilter appname = os.path.basename(__file__) logger = logging.getLogger(appname) def construct_and_run(): def prog(progname): logger.info("doing what %s is doing" % progname) return progname ev = evr.Eventor(name=appname) ev1s = ev.add_event('run_step1') ev2s = ev.add_event('run_step2') ev3s = ev.add_event('run_step3') s1 = ev.add_step('s1', func=prog, kwargs={'progname': 'prog1'}, triggers={evr.StepStatus.success: (ev2s,),}) s2 = ev.add_step('s2', func=prog, kwargs={'progname': 'prog2'}, triggers={evr.StepStatus.success: (ev3s,), }) s3 = ev.add_step('s3', func=prog, kwargs={'progname': 'prog3'},) ev.add_assoc(ev1s, s1) ev.add_assoc(ev2s, s2) ev.add_assoc(ev3s, s3) ev.trigger_event(ev1s, 1) ev.run() ev.close() if __name__ == '__main__': construct_and_run()
The above example with provide the following log output.
[ 2016-11-30 10:07:48,612 ][ INFO ][ Running step s1[1] ][ main.task_wrapper ] [ 2016-11-30 10:07:48,612 ][ INFO ][ Step completed s1[1], status: success, result 'prog1' ][ main.task_wrapper ] [ 2016-11-30 10:07:50,649 ][ INFO ][ Running step s2[1] ][ main.task_wrapper ] [ 2016-11-30 10:07:50,649 ][ INFO ][ Step completed s2[1], status: success, result 'prog2' ][ main.task_wrapper ] [ 2016-11-30 10:07:52,688 ][ INFO ][ Running step s3[1] ][ main.task_wrapper ] [ 2016-11-30 10:07:52,689 ][ INFO ][ Step completed s3[1], status: success, result 'prog3' ][ main.task_wrapper ] [ 2016-11-30 10:07:53,700 ][ INFO ][ Processing finished with: success ][ main.loop_session_start ]Note: actual logging includes hostname and and procname, e.g., "[ mbp02 ][ MainProcess ]". These information was omitted from logging herein.
Eventor (line 10) defines eventor using default database configuration (see bellow).
add_event (e.g., line 12) adds an event named run_step1 to the respective eventor object.
add_step (e.g., line 16) adds step s1 which when triggered would run predefined function prog with key words parameters progname='prog1'. Additionally, when step would end, if successful, it would trigger event evs2
add_assoc (e.g., line 22) links event evs1 and step s1.
trigger_event (line 26) marks event evs1; when triggers, event is associated with sequence. This would allow multiple invocation.
ev() (line 27) invoke Eventor process that would looks for triggers and tasks to act upon. It ends when there is nothing to do.
logger and Eventor names (lines 6 and 14), Eventor is using hierarchical logger based on Eventor name argument. As such naming convention needs to be aligned among all the files participating with the run.
One important artifact used in Eventor is program's runner file. Runner file database (sqlite) will be created at execution, if not directed otherwise, at the location of the run (UNIX's pwd). This file contains information on tasks and triggers that are used in the run and in recovery.
Eventor(name='', store='', run_mode=RUM_RESTART, recovery_run=None, run_id='', config={})
name: string id for Eventor object initiated.
store: Eventor mechanism is built to work with SQLAlchemy. If store is provided, Eventor first check if store is a tag within config under EVENTOR.DATABASE (or whatever the environment variables EVENTOR_CONFIG_TAG and EVENTOR_DB_CONFIG_TAG points to) section. If the tag exists, it will pick its configuration as database configuration. If store is empty, Eventor will try to look for default database configuration. Otherwise, store will be considered as a path to file that would store runnable (sqlite) information; If not provided, calling module path and name will be used with '.db' extension instead of '.py'.
run_mode: can be either RUN_RESTART (default) or RUN_RECOVER; in restart, new instance or the run will be created. In recovery, if shared_db is set, run_id or the recovered program must be provided.
recovery_run: if RUN_RECOVER is used, recovery_run will indicate specific instance of previously recovery run that would be executed.If not provided, latest run would be used.
run_id: unique ID for the program run (excluding recovery_run). It is mandatory in shared_db mode, and if not provided, will be generated.
config: keyword dictionary of default configurations. Available keywords and their default values:
Name Default Value Description workdir /tmp place to create necessary artifacts (not in use) logdir /tmp place to create debug and error log files task_construct mp.Process method to use for execution of steps max_concurrent 1 maximum concurrent processing, if value <1, no limit will be pose stop_on_exception True if an exception occurs in a step, stop all processes. If True, new processes will not start. But running processes will be permitted to finish sleep_between_loops 1 seconds to sleep between iteration of checking triggers and tasks shared_db False if set, db must not be in memory. signals that multiple programs will use the same database tables. envvar_prefix EVENTOR
defined for each step:
   {envvar_prefix}STEP_NAME
   {envvar_prefix}STEP_SEQUENCE
   {envvar_prefix}STEP_RECOVERY
   {envvar_prefix}LOGGER_NAMEssh_config ~/.ssh/config
- SSH configuration file to use with SSH remote
Invocation of steps.
ssh_host SSH host configuration name prime host. ssh_port SSH port to use for SSH connectivity LOGGING dictionary of logging configurations. DATABASES dictionary of database configurations.
EVENTOR: debug: False task_construct: process envvar_prefix: EVENTOR_ max_concurrent: -1 stop_on_exception: True sleep_between_loops: 0.25 sequence_arg_name: None day_to_keep_db: 5 remote_method: ssh pass_logger_to_task: False shared_db: False DATABASES: sqfile1: dialect: sqlite database: /tmp/runly.db pgdb1: dialect: postgresql drivername : psycopg2 username: pgusername password: pgpassword host: ubuntu-guest-02 port: 5433 database: pyground schema: play LOGGING: logging_level: 10 logdir: /var/log/eventor level_formats: 10: ('[ %(asctime)-15s ][ %(host)s ][ %(processName)-11s ][ %(levelname)-7s ]' '[ %(message)s ][ %(module)s.%(funcName)s(%(lineno)d) ]') default: ('[ %(asctime)-15s ][ %(host)s ][ %(processName)-11s ]' '[ %(levelname)-7s ][ %(message)s ]') consolidate: False console: True file_prefix: '' file_suffix: '' file_mode: 'a' maxBytes: 0 backupCount: 0 encoding: 'utf8' delay: False when: 'h' interval: 1 utc: False atTime: 86400
It is possible to create configuration for sqlite memory with cache=shared. However, SQLAlchemy will not work with it well in threaded and multiprocessing environment. Hence, at this point, Eventor does not support it.
add_event(name, expr=None)
name: string unique id for event
- expr: logical expression 'sqlalchemy' style to automatically raise this expression.
syntax:
- if expression is of the first style, logical and will apply.
- the second expression will apply logical or.
- the basic atom in expression is even which is the product of add_event.
Event object to use in other add_event expressions, add_assoc methods, or with add_step triggers.
add_step(name, func, args=(), kwargs={}, triggers={}, acquires=[], releases=None, recovery={}, config={})
name: string unique id for step
func: callable object that would be call at time if step execution
args: tuple of values that will be passed to func at calling
kwargs: keywords arguments that will be passed to func at calling
triggers: mapping of step statuses to set of events to be triggered as in the following table:
status description STEP_READY set when task is ready to run (triggered) STEP_ACTIVE set when task is running STEP_SUCCESS set when task is successful STEP_FAILURE set when task fails STEP_COMPLETE stands for success or failure of task acquires: list of tuples of resource pool and amount of resources to acquire before starting.
releases: list of tuples of resources pool and amount of resources to release once completed. If None, defaults to acquires. If set to empty list, none of the acquired resources would be released.
recovery: mapping of state status to how step should be handled in recovery:
status default description STEP_READY StepReplay.rerun if in recovery and previous status is ready, rerun STEP_ACTIVE StepReplay.rerun if in recovery and previous status is active, rerun STEP_FAILURE StepReplay.rerun if in recovery and previous status is failure, rerun STEP_SUCCESS StepReplay.skip if in recovery and previous status is success, skip config: keywords mapping overrides for step configuration.
name default description stop_on_exception True stop flow if step ends with Exception
Step object to use in add_assoc method.
add_assoc(event, *assocs, delay=0)
event: event objects as provided by add_event.
assocs: list of associations objects. List is composed from either events (as returned by add_event) or steps (as returned by add_step)
delay: seconds to wait, once event is triggered, before engaging its associations
N/A
trigger_event(event, sequence=None)
event: event objects as provided by add_event.
sequence: unique association of triggered event. Event can be triggered only once per sequence. All derivative triggers will carry the same sequence.
N/A
run(max_loops=-1)
when calling run, information is built and loops evaluating events and task starts are executed. In each loop events are raised and tasks are performed. max_loops parameters allows control of how many loops to execute.
In simple example, ev.run() engage Eventor's run() method.
- max_loops: max_loops: number of loops to run. If positive, limits number of loops.
defaults to negative, which would run loops until there are no events to raise and no task to run.
If there was a failure that was not followed by event triggered, result will be False.
close()
when calling close, Eventor object will close its open artifacts. This is similar to close method on multiprocessing Pool.
In simple example, ev.close() engage Eventor's close() method.
N/A.
N/A.
When running in recovery, unless indicated otherwise, latest run (initial or recovery) would be used.
Note that when running a program with the intent to use its recovery capabilities, in-memory store cannot be use. Instead, physical storage must be used.
Here is an example for recovery program and run.
import eventor as evr import logging import math from acrilib import LoggerAddHostFilter appname = os.path.basename(__file__) logger = logging.getLogger(appname) def square(x): y = x*x logger.info("Square of %s is %s" % (x, y)) return y def square_root(x): y = math.sqrt(x) logger.info("Square root of %s is %s" % (x, y)) return y def divide(x, y): z = x/y logger.info("dividing %s by %s is %s" % (x, y, z)) return z def build_flow(run_mode=evr.RUN_RESTART, param=9, run_id=None): ev = evr.Eventor(name=appname, run_mode=run_mode, run_id=run_id, config={'LOGGING': {'logging_level': logging.INFO}},)) ev1s = ev.add_event('run_step1') ev1d = ev.add_event('done_step1') ev2s = ev.add_event('run_step2') ev2d = ev.add_event('done_step2') ev3s = ev.add_event('run_step3', expr=(ev1d, ev2d)) s1 = ev.add_step('s1', func=square, kwargs={'x': 3}, triggers={evr.STEP_SUCCESS: (ev1d, ev2s,)},) s2 = ev.add_step('s2', square_root, kwargs={'x': param}, triggers={evr.STEP_SUCCESS: (ev2d,), }, recovery={evr.STEP_FAILURE: evr.STEP_RERUN, evr.STEP_SUCCESS: evr.STEP_SKIP}) s3 = ev.add_step('s3', divide, kwargs={'x': 9, 'y': 3},) ev.add_assoc(ev1s, s1) ev.add_assoc(ev2s, s2) ev.add_assoc(ev3s, s3) ev.trigger_event(ev1s, 3) return ev def construct_and_run(): # start regularly; it would fail in step 2 ev = build_eventor(param=-9) run_id = ev.run_id ev.run() ev.close() # rerun in recovery ev = build_eventor(evr.RUN_RECOVER, param=9, run_id=run_id) ev.run() ev.close() if __name__ == '__main__': construct_and_run()
[ 2016-12-07 08:37:53,541 ][ INFO ][ Eventor store file: /eventor/example/runly03.run.db ] [ 2016-12-07 08:37:53,586 ][ INFO ][ [ Step s1/3 ] Trying to run ] [ 2016-12-07 08:37:53,588 ][ INFO ][ Square of 3 is 9 ] [ 2016-12-07 08:37:53,588 ][ INFO ][ [ Step s1/3 ] Completed, status: TaskStatus.success ] [ 2016-12-07 08:37:55,644 ][ INFO ][ [ Step s2/3 ] Trying to run ] [ 2016-12-07 08:37:55,647 ][ INFO ][ [ Step s2/3 ] Completed, status: TaskStatus.failure ] [ 2016-12-07 08:37:56,663 ][ ERROR ][ Exception in run_action: <Task(id='2', step_id='s2', sequence='3', recovery='0', pid='8112', status='TaskStatus.failure', created='2016-12-07 14:37:55.625870', updated='2016-12-07 14:37:55.633819')> ] [ 2016-12-07 08:37:56,663 ][ ERROR ][ ValueError('math domain error',) ] [ 2016-12-07 08:37:56,663 ][ ERROR ][ File "/sand/eventor/eventor/main.py", line 62, in task_wrapper result=step(seq_path=task.sequence) File "/sand/eventor/eventor/step.py", line 82, in __call__ result=func(*func_args, **func_kwargs) File "/eventor/example/runly03.py", line 66, in square_root y=math.sqrt(x) ] [ 2016-12-07 08:37:56,663 ][ INFO ][ Stopping running processes ] [ 2016-12-07 08:37:56,667 ][ INFO ][ Processing finished with: failure ] [ 2016-12-07 08:37:56,670 ][ INFO ][ Eventor store file: /eventor/example/runly03.run.db ] [ 2016-12-07 08:37:57,736 ][ INFO ][ [ Step s2/3 ] Trying to run ] [ 2016-12-07 08:37:57,739 ][ INFO ][ Square root of 9 is 3.0 ] [ 2016-12-07 08:37:57,739 ][ INFO ][ [ Step s2/3 ] Completed, status: TaskStatus.success ] [ 2016-12-07 08:38:00,798 ][ INFO ][ [ Step s3/3 ] Trying to run ] [ 2016-12-07 08:38:00,800 ][ INFO ][ dividing 9 by 3 is 3.0 ] [ 2016-12-07 08:38:00,800 ][ INFO ][ [ Step s3/3 ] Completed, status: TaskStatus.success ] [ 2016-12-07 08:38:01,824 ][ INFO ][ Processing finished with: success ]
The function build_flow (code line 24) build an Eventor flow using three functions defined in advance. Since no specific store is provided in Eventor instantiation, a default runner store is assigned (code line 25). In this build, step s2 (lines 30-35) is being set with recovery directives.
The first build and run is done in lines 47-48. In this run, a parameter that would cause the second step to fail is being passed. As a result, flow fails. Output lines 1-17 is associated with the first run.
The second build and run is then initiated. In this run, parameter is set to a value that would pass step s2 and run mode is set to recovery (code lines 51-52). Eventor skips successful steps and start executing from failed steps onwards. Output lines 18-25 reflects successful second run.
There are situations in which it is desire to hold off activating a task. This behavior is captured in Eventor as a delayed association.
Associations can be made delayed. Assuming source event is associated to target event with time delay. When source event is triggered, Eventor will wait time delay seconds before triggering target event.
In such situations, it sometimes desire to run Eventor engine in specific period on a time line instead of continuously. For example, if Eventor is synchronizing activities that has 6 hours association delay. Instead of running Eventor continuously, it can be set to run every 5 minutes, and save computing resources on the side.
With delayed associations, Eventor can run in continue run mode (RunMode.continue_). When running in continue, Eventor will pick up from where it left last run.
The following example present delayed association with continue run mode.
import eventor as evr import logging import os import time appname = os.path.basename(__file__) logger = logging.getLogger(appname) def prog(progname): logger.info("doing what %s is doing" % progname) logger.info("EVENTOR_STEP_SEQUENCE: %s" % os.getenv("EVENTOR_STEP_SEQUENCE")) return progname def build_flow(run_mode): ev = evr.Eventor(name=appname, run_mode=run_mode,) ev1s = ev.add_event('run_step1') ev2s = ev.add_event('run_step2') ev3s = ev.add_event('run_step3') s1 = ev.add_step('s1', func=prog, kwargs={'progname': 'prog1'}, triggers={evr.STEP_SUCCESS: (ev2s,)}) s2 = ev.add_step('s2', func=prog, kwargs={'progname': 'prog2'}, triggers={evr.STEP_SUCCESS: (ev3s,)}) s3 = ev.add_step('s3', func=prog, kwargs={'progname': 'prog3'},) ev.add_assoc(ev1s, s1, delay=0) ev.add_assoc(ev2s, s2, delay=10) ev.add_assoc(ev3s, s3, delay=10) ev.trigger_event(ev1s, 1) return ev def construct_and_run(): ev = build_flow(run_mode=evr.RUN_RESTART) ev.run() ev.close() if __name__ == '__main__': construct_and_run()
[ 2017-08-16,16:31:29.277048 ][ Task-s1(1) ][ INFO ][ [ Step s1/1 ] Trying to run ] [ 2017-08-16,16:31:29.277903 ][ Task-s1(1) ][ INFO ][ doing what prog1 is doing ] [ 2017-08-16,16:31:29.278114 ][ Task-s1(1) ][ INFO ][ EVENTOR_STEP_SEQUENCE: 1 ] [ 2017-08-16,16:31:29.278360 ][ Task-s1(1) ][ INFO ][ [ Step s1/1 ] Completed, status: TaskStatus.success ] [ 2017-08-16,16:31:41.028196 ][ Task-s2(1) ][ INFO ][ [ Step s2/1 ] Trying to run ] [ 2017-08-16,16:31:41.029191 ][ Task-s2(1) ][ INFO ][ doing what prog2 is doing ] [ 2017-08-16,16:31:41.029429 ][ Task-s2(1) ][ INFO ][ EVENTOR_STEP_SEQUENCE: 1 ] [ 2017-08-16,16:31:41.029697 ][ Task-s2(1) ][ INFO ][ [ Step s2/1 ] Completed, status: TaskStatus.success ] [ 2017-08-16,16:32:02.931265 ][ Task-s3(1) ][ INFO ][ [ Step s3/1 ] Trying to run ] [ 2017-08-16,16:32:02.932407 ][ Task-s3(1) ][ INFO ][ doing what prog3 is doing ] [ 2017-08-16,16:32:02.932661 ][ Task-s3(1) ][ INFO ][ EVENTOR_STEP_SEQUENCE: 1 ] [ 2017-08-16,16:32:02.932940 ][ Task-s3(1) ][ INFO ][ [ Step s3/1 ] Completed, status: TaskStatus.success ] [ 2017-08-16,16:32:03.014584 ][ MainProcess ][ INFO ][ Processing finished with: success; outstanding tasks: 0 ]
The example program builds and runs Eventor sequence 4 times. The build involves three tasks that would run sequentially. They are associated to each other with delay of 10 seconds each (lines 26 and 28.)
The first time, sequence is build with restart run mode (line 35). In this case, the sequence is initiated. The next four runs are in continue run mode (line 48). Each of those run continue its preceding run. To have it show the point, a varying delay is introduced between runs (lines 46-47).
Each run limits the number of loop to a single loop (lines 40 and 50). A single loop entails Eventor executing triggers and tasks until there is none to execute. It may be though that there are still outstanding delayed association to act upon.
This behavior is different than continuous run (using max_loops=-1), which is the default. In such run, Eventor will continue to loop until there are no triggers, tasks, and delayed association to process.
Eventor runs can be observed in example output lines 1-5, 6, 7-11, 12, and 13-17 each. Note that the second and forth runs had not trigger to execute on. The associated tasks' delays was not yet matured.
add_step allows association of step with resources. If acquires argument is provided, before step starts, Eventor will attempt to reserve resources. Step will be executed only when resources are secured.
When release argument is provided, resources resources listed as its value will be released when step is done. If release is None, whatever resources stated by acquires would be released. If the empty list is set as value, no resource would be released.
To use resources, program to use Resource and ResourcePool from acris.virtual_resource_pool. Example for such definitions are below.
import eventor as evr from acris import virtual_resource_pool as vrp class Resources1(vrp.Resource): pass class Resources2(vrp.Resource): pass rp1 = vrp.ResourcePool('RP1', resource_cls=Resources1, policy={'resource_limit': 2, }).load() rp2 = vrp.ResourcePool('RP2', resource_cls=Resources2, policy={'resource_limit': 2, }).load() ev = evr.Eventor() s1 = ev.add_step('s0.s00.s1', func=prog, kwargs={'progname': 'prog1'}, acquires=[(rp2, 1), ],)
Eventor program can work in a clustered environment. In this arrangement, steps can be defined to run on different nodes in the cluster. This is possible granted:
- SSH is defined among cluster nodes.
- Eventor DB is shared among cluster nodes.
- Program environment is the seamlessly-the-same among cluster nodes.
Eventor will be launched from one host, server. It will then start the same program on every associated host relevant to program, clients. Client programs will skip starting steps (steps with no )
When working on distributed environment, Eventor assumes that ssh is set properly among participating hosts.
To allow ssh run command with .profile (or .bash_profile) are not automatically executed, add the following before RSA key in .ssh/authorizedkeys
command "if [[ \"x${SSH_ORIGINAL_COMMAND}x\" != \"xx\" ]]; then source ~/.profile; eval \"${SSH_ORIGINAL_COMMAND}\"; else /bin/bash --login; fi;" <key>
Eventor program would be launched on all cluster nodes relevant to the program.
The following is some of the major tasks intended to be completed into this product.
- asynchronous tasks: embed mechanism to launch asynchronous tasks.
- remote callback mechanisms: allow remote asynchronous tasks communicate with Eventor (TCP/IP, HTTP, etc.)
- virtual resources shared across distributed environment.
- improve SSHPipe to better indicate SSH channel was established.
- added database configuration allowing the use of SqlAlchemy database engines.
- added shared_db to indicate db is shared among multiple programs and runs.
- added run_id as unique identifier for program run (not to be confused with recovery).
- improved documentation to reflect the need for mp.freeze_support() and mp.set_start_method('spawn').
- added dependency on namedlist, and PyYAML, packages.
- bug fix in delay.
- SSH remote invocation of steps.
- use of socket based logging.
- centralized logging of remote agents in prime server.
- Added {envvar_prefix}LOGGER_NAME to allow step logger to be set appropriately.
- Eventor github project (Eventor github project) has additional examples with more complicated flows.
- SSH Pipe blog clarifying the mechanism using by Eventor for remote steps and centralized logging.