Class: Datadog::Tracing::Workers::AsyncTraceWriter
- Inherits:
-
TraceWriter
- Object
- Core::Worker
- TraceWriter
- Datadog::Tracing::Workers::AsyncTraceWriter
- Includes:
- Core::Workers::Polling, Core::Workers::Queue
- Defined in:
- lib/datadog/tracing/workers/trace_writer.rb
Overview
Writes traces to transport asynchronously, using a thread & buffer.
Constant Summary collapse
- DEFAULT_BUFFER_MAX_SIZE =
1000
- FORK_POLICY_ASYNC =
:async
- FORK_POLICY_SYNC =
:sync
Constants included from Core::Workers::Polling
Core::Workers::Polling::DEFAULT_SHUTDOWN_TIMEOUT
Instance Attribute Summary collapse
-
#async ⇒ Object
writeonly
Sets the attribute async.
Attributes included from Core::Workers::Queue
Attributes inherited from TraceWriter
Attributes inherited from Core::Worker
Instance Method Summary collapse
- #after_fork ⇒ Object
- #async? ⇒ Boolean
- #dequeue ⇒ Object
- #enqueue(trace) ⇒ Object
- #fork_policy=(policy) ⇒ Object
-
#initialize(options = {}) ⇒ AsyncTraceWriter
constructor
A new instance of AsyncTraceWriter.
-
#perform(traces) ⇒ Object
NOTE: #perform is wrapped by other modules: Polling –> Async –> IntervalLoop –> AsyncTraceWriter –> TraceWriter.
- #stop(force_stop = false, timeout = @shutdown_timeout) ⇒ Object
-
#work_pending? ⇒ Boolean
Are there more traces to be processed next?.
-
#write(trace) ⇒ Object
WARNING: This method breaks the Liskov Substitution Principle – TraceWriter#write is spec’d to return the result from the writer, whereas this method returns something else when running in async mode.
Methods included from Core::Workers::Polling
#enabled=, #enabled?, included
Methods included from Core::Workers::Queue
Methods inherited from TraceWriter
#flush_completed, #flush_traces, #process_traces, #write_traces
Constructor Details
#initialize(options = {}) ⇒ AsyncTraceWriter
Returns a new instance of AsyncTraceWriter.
90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 |
# File 'lib/datadog/tracing/workers/trace_writer.rb', line 90 def initialize( = {}) # Workers::TraceWriter settings super # Workers::Polling settings self.enabled = .fetch(:enabled, true) # Workers::Async::Thread settings @async = true self.fork_policy = .fetch(:fork_policy, FORK_POLICY_ASYNC) # Workers::IntervalLoop settings self.loop_base_interval = [:interval] if .key?(:interval) self.loop_back_off_ratio = [:back_off_ratio] if .key?(:back_off_ratio) self.loop_back_off_max = [:back_off_max] if .key?(:back_off_max) # Workers::Queue settings @buffer_size = .fetch(:buffer_size, DEFAULT_BUFFER_MAX_SIZE) self.buffer = TraceBuffer.new(@buffer_size) @shutdown_timeout = .fetch(:shutdown_timeout, Core::Workers::Polling::DEFAULT_SHUTDOWN_TIMEOUT) end |
Instance Attribute Details
#async=(value) ⇒ Object (writeonly)
Sets the attribute async
87 88 89 |
# File 'lib/datadog/tracing/workers/trace_writer.rb', line 87 def async=(value) @async = value end |
Instance Method Details
#after_fork ⇒ Object
170 171 172 173 174 175 176 177 178 179 |
# File 'lib/datadog/tracing/workers/trace_writer.rb', line 170 def after_fork # In multiprocess environments, forks will share the same buffer until its written to. # A.K.A. copy-on-write. We don't want forks to write traces generated from another process. # Instead, we reset it after the fork. (Make sure any enqueue operations happen after this.) self.buffer = TraceBuffer.new(@buffer_size) # Switch to synchronous mode if configured to do so. # In some cases synchronous writing is preferred because the fork will be short lived. @async = false if @writer_fork_policy == FORK_POLICY_SYNC end |
#async? ⇒ Boolean
146 147 148 |
# File 'lib/datadog/tracing/workers/trace_writer.rb', line 146 def async? @async == true end |
#dequeue ⇒ Object
135 136 137 138 139 |
# File 'lib/datadog/tracing/workers/trace_writer.rb', line 135 def dequeue # Wrap results in Array because they are # splatted as args against TraceWriter#perform. [buffer.pop] end |
#enqueue(trace) ⇒ Object
131 132 133 |
# File 'lib/datadog/tracing/workers/trace_writer.rb', line 131 def enqueue(trace) buffer.push(trace) end |
#fork_policy=(policy) ⇒ Object
150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 |
# File 'lib/datadog/tracing/workers/trace_writer.rb', line 150 def fork_policy=(policy) # Translate to Workers::Async::Thread policy thread_fork_policy = case policy when Core::Workers::Async::Thread::FORK_POLICY_STOP policy when FORK_POLICY_SYNC # Stop the async thread because the writer # will bypass and run synchronously. Core::Workers::Async::Thread::FORK_POLICY_STOP else Core::Workers::Async::Thread::FORK_POLICY_RESTART end # Update thread fork policy super(thread_fork_policy) # Update local policy @writer_fork_policy = policy end |
#perform(traces) ⇒ Object
NOTE: #perform is wrapped by other modules:
Polling --> Async --> IntervalLoop --> AsyncTraceWriter --> TraceWriter
WARNING: This method breaks the Liskov Substitution Principle – TraceWriter#perform is spec’d to return the result from the writer, whereas this method always returns nil.
118 119 120 121 122 123 124 |
# File 'lib/datadog/tracing/workers/trace_writer.rb', line 118 def perform(traces) super(traces).tap do |responses| loop_back_off! if responses.find(&:server_error?) end nil end |
#stop(force_stop = false, timeout = @shutdown_timeout) ⇒ Object
126 127 128 129 |
# File 'lib/datadog/tracing/workers/trace_writer.rb', line 126 def stop(force_stop = false, timeout = @shutdown_timeout) buffer.close if running? super end |
#work_pending? ⇒ Boolean
Are there more traces to be processed next?
142 143 144 |
# File 'lib/datadog/tracing/workers/trace_writer.rb', line 142 def work_pending? !buffer.empty? end |
#write(trace) ⇒ Object
WARNING: This method breaks the Liskov Substitution Principle – TraceWriter#write is spec’d to return the result from the writer, whereas this method returns something else when running in async mode.
183 184 185 186 187 188 189 190 191 192 |
# File 'lib/datadog/tracing/workers/trace_writer.rb', line 183 def write(trace) # Start worker thread. If the process has forked, it will trigger #after_fork to # reconfigure the worker accordingly. # NOTE: It's important we do this before queuing or it will drop the current trace, # because #after_fork resets the buffer. perform # Queue the trace if running asynchronously, otherwise short-circuit and write it directly. async? ? enqueue(trace) : write_traces([trace]) end |