Class: Gitlab::StackProf
- Inherits:
-
Object
- Object
- Gitlab::StackProf
- Defined in:
- lib/gitlab/stack_prof.rb
Constant Summary collapse
- DEFAULT_FILE_PREFIX =
Dir.tmpdir
- DEFAULT_TIMEOUT_SEC =
30
- DEFAULT_MODE =
:cpu
- DEFAULT_INTERVAL_US =
Sample interval as a frequency in microseconds (~99hz); appropriate for CPU profiles
10_100
- DEFAULT_INTERVAL_EVENTS =
Sample interval in event occurrences (n = every nth event); appropriate for allocation profiles
100
Class Method Summary collapse
-
.install ⇒ Object
this is a workaround for sidekiq, which defines its own SIGUSR2 handler.
- .interval(mode) ⇒ Object
- .log_event(event, labels = {}) ⇒ Object
- .on_worker_start ⇒ Object
Class Method Details
.install ⇒ Object
this is a workaround for sidekiq, which defines its own SIGUSR2 handler. by defering to the sidekiq startup event, we get to set up our own handler late enough. see also: github.com/mperham/sidekiq/pull/4653
21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 |
# File 'lib/gitlab/stack_prof.rb', line 21 def self.install require 'stackprof' require 'tmpdir' if Gitlab::Runtime.sidekiq? Sidekiq.configure_server do |config| config.on :startup do on_worker_start end end else Gitlab::Cluster::LifecycleEvents.on_worker_start do on_worker_start end end end |
.interval(mode) ⇒ Object
135 136 137 |
# File 'lib/gitlab/stack_prof.rb', line 135 def self.interval(mode) mode == :object ? DEFAULT_INTERVAL_EVENTS : DEFAULT_INTERVAL_US end |
.log_event(event, labels = {}) ⇒ Object
127 128 129 130 131 132 133 |
# File 'lib/gitlab/stack_prof.rb', line 127 def self.log_event(event, labels = {}) Gitlab::AppJsonLogger.info({ event: 'stackprof', message: event, pid: Process.pid }.merge(labels.compact)) end |
.on_worker_start ⇒ Object
38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 |
# File 'lib/gitlab/stack_prof.rb', line 38 def self.on_worker_start log_event('listening for SIGUSR2 signal') # create a pipe in order to propagate signal out of the signal handler # see also: https://cr.yp.to/docs/selfpipe.html read, write = IO.pipe # create a separate thread that polls for signals on the pipe. # # this way we do not execute in signal handler context, which # lifts restrictions and also serializes the calls in a thread-safe # manner. # # it's very similar to a goroutine and channel design. # # another nice benefit of this method is that we can timeout the # IO.select call, allowing the profile to automatically stop after # a given interval (by default 30 seconds), avoiding unbounded memory # growth from a profile that was started and never stopped. t = Thread.new do timeout_s = ENV['STACKPROF_TIMEOUT_S']&.to_i || DEFAULT_TIMEOUT_SEC current_timeout_s = nil loop do read.getbyte if IO.select([read], nil, nil, current_timeout_s) if ::StackProf.running? stackprof_file_prefix = ENV['STACKPROF_FILE_PREFIX'] || DEFAULT_FILE_PREFIX stackprof_out_file = "#{stackprof_file_prefix}/stackprof.#{Process.pid}.#{SecureRandom.hex(6)}.profile" log_event( 'stopping profile', profile_filename: stackprof_out_file, profile_timeout_s: timeout_s ) ::StackProf.stop ::StackProf.results(stackprof_out_file) current_timeout_s = nil else mode = ENV['STACKPROF_MODE']&.to_sym || DEFAULT_MODE stackprof_interval = ENV['STACKPROF_INTERVAL']&.to_i stackprof_interval ||= interval(mode) log_event( 'starting profile', profile_mode: mode, profile_interval: stackprof_interval, profile_timeout: timeout_s ) ::StackProf.start( mode: mode, raw: Gitlab::Utils.to_boolean(ENV['STACKPROF_RAW'] || 'true'), interval: stackprof_interval ) current_timeout_s = timeout_s end end rescue StandardError => e log_event("stackprof failed: #{e}") end t.abort_on_exception = true # in the case of puma, this will override the existing SIGUSR2 signal handler # that can be used to trigger a restart. # # puma cluster has two types of restarts: # * SIGUSR1: phased restart # * SIGUSR2: restart # # phased restart is not supported in our configuration, because we use # preload_app. this means we will always perform a normal restart. # additionally, phased restart is not supported when sending a SIGUSR2 # directly to a puma worker (as opposed to the master process). # # the result is that the behaviour of SIGUSR1 and SIGUSR2 is identical in # our configuration, and we can always use a SIGUSR1 to perform a restart. # # thus, it is acceptable for us to re-appropriate the SIGUSR2 signal, and # override the puma behaviour. # # see also: # * https://github.com/puma/puma/blob/master/docs/signals.md#puma-signals # * https://github.com/mperham/sidekiq/wiki/Signals Signal.trap('SIGUSR2') do write.write('.') end end |