package org.arbeitspferde.groningen.proto
Get desktop application:
View/edit binary Protocol Buffers messages
message
event.proto:47optional uint64 experiment_id = 1
The ID for this subject.
optional uint64 time = 2
The time the event took place.
optional string groningen_user = 3
The MDB user of the running Groningen.
optional string groningen_serving_address = 4
The serving address of the host running Groningen: ${hostname}:${port}.
optional string subject_user = 5
The user of the running subject.
optional string subject_serving_address = 6
The serving address of the experimental subject.
optional EventEntry.Type type = 7
The event type.
repeated jvm_flag = 8
The running JVM flags.
repeated score = 9
The fitness score associated with the event.
optional groningen_configuration = 10
The configuration parameters that Groningen was operating under when the event was recorded.
repeated pause_event = 11
An inventory, if any, of the pause time events.
optional uint64 groningen_start_time = 12
The time at which the Groningen instance started. This is used for fingerprinting.
message
event.proto:76Representation of fitness score.
Used in:
required string name = 1
The name of the fitness score.
required float score = 2
The assigned value of the fitness score.
required float coefficient = 3
The coefficient for the score.
message
event.proto:65Representation of Groningen-managed JVM flags.
Used in:
required string name = 1
The name of the flag.
required string value = 2
The flag's value.
required bool managed_by_groningen = 3
Whether this flag was explicitly set by Groningen or was inherited from subject.
message
event.proto:88Representation of pause-time metrics.
Used in:
required double duration_in_seconds = 1
The duration of the event.
What type of event was observed.
Used in:
START = 0
A subject has been started under a new set of experimental parameters.
UNEXPECTED_DEATH = 1
A subject was found by Groningen mid-way during experimentation unexpectedly.
EXPERIMENT_END = 2
A subject's experiment was terminated by Groningen.
message
exp_arg_file.proto:27A bundle of the experimental args and anything else the injection script needs to do its jobs
optional string args = 1
the experimental args themselves
optional string master_serving_address = 2
the backpointer back to the instance that is controlling the subjects.
repeated string tuned_arg = 3
list of all command line args that Groningen is actively tuning which can be used to produce of list of args in the default set that Groningen is not tuning
optional string pipeline_id = 4
message
event.proto:162optional string fingerprint_of_unique_subject_identifier = 1
optional string fingerprint_of_unique_experiment_identifier = 2
optional uint64 experiment_id = 3
message
event.proto:152repeated score = 1
optional float total_score = 2
message
event.proto:38Representation of fitness score.
Used in:
optional string name = 1
The name of the fitness score.
optional float score = 2
The assigned value of the fitness score.
optional float coefficient = 3
The coefficient for the score.
message
params.proto:30Used in:
optional string input_log_name = 1
The basename of the subjects log file that Groningen will parse as input
optional int32 duration = 2
time in minutes that the experiments run in production
optional int32 restart = 3
experimental subject restart threshold used by the Executor and Validator
optional double latency_weight = 5
Weight of the latency score within the hypothesizer
optional double throughput_weight = 6
Weight of the throughput score within the hypothesizer
optional double memory_weight = 8
Weight of the memory score within the hypothesizer
optional int32 num_crossovers = 9
Number cross-over points for mating within hypothesizer
optional double mutation_prob = 10
Mutation probability for the offsprings
optional int32 stagnant_gens = 11
Number of stagnant generations before GA terminates. The value 0 disables this termination condition check.
optional int32 elite_count = 12
Number of elite individuals in the populations for the IncrementalEvolutionEngine NOTE: is not updated after the initial pass through the loop.
optional double pause_time_latency_score_percentile = 13
The percentile used to compute the latency score (A value from 0 to 100)
optional int32 number_of_executor_threads = 14
The number of threads that the Executor uses to monitor experimental subjects. Quite likely, you'll want to increase this as you increase the total number of subjects in your experiment. Less threads probably implies alonger period of time between sucessive probes of a given experimental subject.
optional GroningenParams.PipelineSynchMode pipeline_sync_type = 15
optional int32 health_check_timeout = 16
How long we wait for a health check check
optional int32 health_check_poll_interval = 17
How long we sleep between health checks.
optional string health_check_server_type = 18
The type of server that is supposed to be registered at the recipient end.
optional int32 metric_query_deadline_secs = 19
The default deadline in seconds for metric queries.
optional int32 metric_query_channel_shutdown_delay_secs = 20
The default channel TTL in seconds for metric query channel.
optional int32 metric_query_retry_maximum_count = 21
The maximum number of times the metric query pipeline will attempt to query.
optional int32 metric_query_retry_delay_secs = 22
The initial delay in seconds between retries for metric query failures.
optional int32 executor_wait_for_one_subject_restart_ms = 23
Deadline (ms) of how long to wait for a subject to restart.
optional int32 executor_sleep_btwn_polling_metrics_ms = 24
Interval (ms) we wait between an iteration of polling subjects
optional int32 extractor_number_of_extractor_threads = 25
The number of extractor threads.
optional int32 maximum_inflight_subject_restart_count = 26
Maximum number of subjects to restart at once.
optional int32 subject_manipulation_deadline_ms = 27
The deadline (ms) for subject manipulation to occur.
optional string deprecated_field_a = 28
File that contains the last experiment.
optional int32 default_in_memory_cache_ttl = 29
The maximum TTL in seconds for objects in the memory-sensitive caches.
optional int32 max_subjects_in_flight_fraction = 30
Divisor for max subjects in-flight for rolling subject group restart. (maxInFlight = # subjects / this divisor)
optional int32 subject_group_number_of_shuffles = 31
The number of times to shuffle the subjects.
optional int32 subjects_to_display = 32
The number of subjects to display in Groningen's HUD.
optional string event_log_prefix = 33
The path along with base string prefix for the Groningen event log.
optional int32 event_log_rotate_size_bytes = 34
The quantity in bytes that the Groningen event log may grow before being rotated.
optional int32 event_log_flush_interval_secs = 35
The number of seconds that may transpire between Groningen event log flushing.
optional string additional_exec_paths = 36
Additional paths where executable permissions will be set when running experiments.
optional string requested_pipeline_id = 37
Request specific pipeline ID, otherwise it will be auto generated.
Groningen can run with various amounts of interaction between pipeline steps This value is only read on creation of the pipeline (and not updated between iterations
Used in:
NONE = 0
No synchronization is performed. Groningen will run the subjects for the duration specified in param_block, and on completion of an iteration of the pipeline, will immediately go into another iteration
BASIC_SEMAPHORE = 1
Provides two points of synchronization: - A point at the beginning of an iteration through the pipeline, Groningen waits before the Hypothesizer is run and blocks the rpc until the experimental settings have been generated and the subjects have been restarted in order to incorporate the experimental settings - Another point that signals the experiment's duration is complete and blocks until all gc logs have been retrieved from remote systems, which would allow the experimental subjects to be downed. The duration setting above becomes a hard limit to prevent Groningen from waiting forever should the external agent crash or otherwise fail to signal the end of the experiment.
ITERATION_FINALIZATION_ONLY = 2
Provides a single point of synchronization at the end of an iteration at which the pipeline will pause a flag is set allowing it to continue. The flag can be set before the pipeline reaches the sync point.
message
event.proto:27Representation of Groningen-managed JVM flags.
Used in:
optional string name = 1
The name of the flag.
optional string value = 2
The flag's value.
optional bool managed_by_groningen = 3
Whether this flag was explicitly set by Groningen or was inherited from subject.
message
event.proto:157repeated flag = 1
optional string complete_commandline = 2
message
event.proto:142optional float minimum_duration = 1
optional float maximum_duration = 2
optional float mean_duration = 3
optional float median_duration = 4
optional float standard_deviation_of_durations = 5
optional float total_duration = 6
optional uint64 event_count = 7
message
groningen_config.proto:42Overall configuration message definition While many fields are defined as optional in order to allow default values to be specified at the appropriate level (and no base default defined if such a value is not necessary), many must be defined at some level in the message hierachary as is noted below.
Used in: , ,
repeated cluster = 1
list of cluster config - must not be
optional string user = 16
empty
highest level default user defintion
optional jvm_search_restriction = 18
optional deprecated_message_a = 19
allow restricting the argument search space at the top level
optional param_block = 20
optional int32 number_of_subjects = 21
the parameter structure that will be passed around. Abstracted out to allow easy addition of new variables without changing java interfaces
optional int32 subject_warmup_timeout = 22
Restrict the experiment to the first N subjects (0 to N-1) Default value of 0 implies all subjects in a group are under Groningens's control
The maximum number of seconds that an experimental subject is allowed to _not_ return OK. This time is added to the experiment duration so that subjects are allowed to warmup.
message
groningen_config.proto:128Definition for Cluster Configuration. This sets values for anything within a cluster which will act as defaults for any inner scoped message. Values that appears within inner messages will override values scoped in outer messages.
Used in:
repeated subject_group = 1
optional string cluster = 16
cluster name - must be specified
optional string user = 17
somewhat optional - must be
optional int32 number_of_subjects = 19
specified somewhere in the hierarchy
optional int32 subject_warmup_timeout = 20
Restrict the experiment to the first N subjects (0 to N-1) Default value of 0 implies all subjects in a group are under Groningens's control
optional int32 number_of_default_subjects = 21
The maximum number of seconds that an experimental subject is allowed to _not_ return OK on health probes when running an experiment. Health checking is performed on subjects that are erroring during routine inspections as well as at the beginning of an experiment following the point when all of the subjects are restarted in the Executor stage.
The number of subjects with default settings. These default subjects are scored concurrently with modified subjects in the experiment, but are otherwise untouched by tuning pipeline stages such as the Hypothesizer, Generator and Validator. Using default subjects reduces the number of subjects available for experimentation, however using default subjects provides a way to directly compare the performance of modified settings with unmodified settings. The number of default subjects should not exceed the total number of subjects in the experiment.
message
groningen_config.proto:131Subject Group scoped values
Used in:
repeated subject_config = 1
optional string subject_group_name = 16
Groningen will use all subjects in a group regardless of the number of subjects specified (requiring groups to split app under test to a different subject group) Any under specified subject will have values chosen from random.
subject group name must be
optional string user = 17
specified in order to attach to a set of subjects within it.
somewhat optional - must be
optional string exp_settings_files_dir = 18
specified somewhere in the hierarchy
optional int32 number_of_subjects = 19
Experiments settings used by jvm_arg_injector for this group will be stored in this directory. There will be one file for every subject with the full path looking like this: <exp_settings_files_dir>/<subject index>
optional int32 subject_warmup_timeout = 20
Restrict the experiment to the first N subject (0 to N-1) Default value of 0 implies all subjects in a subject group are under Groningen's control.
repeated string restart_command = 21
The maximum number of seconds that an experimental subject is allowed to _not_ return OK.
optional int32 number_of_default_subjects = 22
The command that should be used to restart this subject group. Overrides the default cluster implementation command.
The number of subjects with default settings. These default subjects are scored concurrently with modified subjects in the experiment, but are otherwise untouched by tuning pipeline stages such as the Hypothesizer, Generator and Validator. Using default subjects reduces the number of subjects available for experimentation, however using default subjects provides a way to directly compare the performance of modified settings with unmodified settings. The number of default subjects should not exceed the total number of subjects in the experiment.
message
groningen_config.proto:134Subject-specific values
Used in:
optional int32 subject_index = 16
optional subject index for the group
optional jvm_parameters = 17
At the subject level, you must define the all arguments if field is included
message
groningen_config.proto:119Used in:
optional string deprecated_a = 16
repeated string deprecated_b = 15
message
groningen_config.proto:50The search space can be manipulated in two ways: - it can be restricted by pinning argument values to a value or a range on a per argument basis - a set of mutations can be fully specified to use as seed points for exploring the space. What follows is a uniform structure for describing both.
Used in: ,
repeated JvmSearchSpace.GcMode gc_mode = 1
At some point in the future, these should be migrated to generic MessageSet based arguments which would be completely extensible. This was not done in mach 1 as most of the commandline abstraction is already built around the CommandLineArg abstraction - redesigning this would likely not allow us to finish the prototype within the task force. The generic MessageSet can be phased in later - either as an additional to existing arguments, or as a complete replacement. NOTE: these fields do not follow the same numbering scheme as the rest of the file as we expect to burn the range.
this is a series so that multiple modes
optional adaptive_size_decrement_scale_factor = 2
can be searched over, though they are mutually exclusive on a given cmd line
optional cms_exp_avg_factor = 3
optional cms_incremental_duty_cycle = 4
optional cms_incremental_duty_cycle_min = 5
optional cms_incremental_offset = 6
optional cms_incremental_safety_factor = 7
optional cms_initiating_occupancy_fraction = 8
optional gc_time_ratio = 9
optional max_gc_pause_millis = 10
optional max_heap_free_ratio = 11
optional max_new_size = 12
optional max_perm_size = 13
optional min_heap_free_ratio = 14
optional new_ratio = 15
optional new_size = 16
optional parallel_gc_threads = 17
optional survivor_ratio = 18
optional tenured_generation_size_increment = 19
optional heap_size = 20
optional young_generation_size_increment = 21
optional bool cms_incremental_mode = 22
optional bool cms_incremental_pacing = 23
optional bool use_cms_initiating_occupancy_only = 24
optional soft_ref_lru_policy_ms_per_mb = 25
The GC modes are mutally exclusive, so capture that fact in an enum
Used in:
USE_CONC_MARK_SWEEP = 0
USE_PARALLEL = 1
USE_PARALLEL_OLD = 2
USE_SERIAL = 3
message
groningen_config.proto:66Hold a specific value or a range of int64s. not following the interaction specified below will result in exceptions... This model was chosen to reduce overall space per subject should one need to specify 100+ subjects (though it would have helped to reduce the size of the arg names as well) with regard to underlying storage limitations.
Used in:
optional int64 value = 1
Specific value to use - mutually
optional int64 floor = 2
exclusive to those below
lower bound (inclusive) for
optional int64 ceiling = 3
range - msg must include ceiling and not value field
upper bound (inclusive) for
optional int64 step_size = 4
range - msg must include floor and not value field
Step size by which the Hypothesizer
message
event.proto:136optional uint64 experiment_id = 1
optional uint64 groningen_start_time = 2
optional string groningen_serving_address = 3
message
event.proto:129The UniqueSubjectIdentifier and UniqueExperimentIdentifier messages below are not used internally in Groningen but rather by our Sawzall event logging processing pipeline. The had originally been defined as ad hoc tuples in the Sawzall script; but because they will be useful for ex post facto data collation and analysis, they have been defined here for anyone to use. Within the context of the Sawzall script, they are used for fingerprinting to generate globally unique identifiers across our data sets.
optional uint64 experiment_id = 1
optional uint64 event_time = 2
optional string groningen_serving_address = 3
optional string subject_serving_address = 4