5.18.1

This is released as part of the Datadog Agent. See Datadog Agent ChangeLog for info about the other components of the agent.

Bugfixes

• Fix a deadlock in stats reporting from the agent #321
• Fix documentation examples on trace.ignore #322
• Fix Windows/Unix rake build tasks #323

5.18.0

This is released as part of the Datadog Agent. See Datadog Agent ChangeLog for info about the other components of the agent.

Improvements

• Add a client-side timeout between agent and API. Default 10 seconds ( #315 )

Bugfixes

• Tag the span with the full sql query even on parsing errors ( #312)

5.17.0 / 08-28-2017

This is released as part of the Datadog Agent. See Datadog Agent ChangeLog for info about the other components of the agent.

Improvements

• Apdex is coming to Datadog! Error duration distribution are added to the stats, which is required for proper Apdex computation #304
• Filter out traces based on their resource name #309

Bugfixes

• Fix issue when re-grouping grouped values in SQL #307

5.14.0 / 06-05-2017

This is released as part of the Datadog Agent. See Datadog Agent ChangeLog for info about the other components of the agent.

https://github.com/DataDog/datadog-trace-agent/milestone/5?closed=1

Improvements

• Agent now pre-sample (drop payloads) when the CPU usage is too high (50% by default, configurable) #240
• Use throttling for all Agent logs #273
• Agent now properly build and run on Windows #275
• Agent now detect top-level name of services #268 #280 #283 #285 #286
• Report per http status code statistics #288
• Improvements to the CI and tests #269 #270 #278

Bugfixes

• Fix support of Java/JRuby payloads #190
• Fix log reporting when the Agent crashes #287
• Fix sublayers stats when combined with client sampling #281

Breaking changes

• Drop legacy port 7777 #271
• Drop unofficial feature of multiple API endpoints #279

5.13.0 / 04-24-2017

This is released as part of the Datadog Agent. See Datadog Agent ChangeLog for info about the other components of the agent.

Improvements

• Allow the trace-agent to run by default as part of a datadog-agent install #260

Bugfixes

• Only accept strictly positive Span durations #252
• Discard traces with duplicate Span IDs #261

5.12.3 / 03-31-2017

This is released as part of the Datadog Agent. See Datadog Agent ChangeLog for info about the other components of the agent.

Improvements

• Added a way to weigh and merge distributions summaries . This has no direct impact on agent usage yet. #253

Bugfixes

• Agent would die and stop processing data when encountering a transient error from our hosted APIs. Fixed by #257
• In some cases, encoded numeric values sent from client to agent were badly encoded/encoded, yielding erroneous statistics. Fixed by #253

5.12.1 / 03-24-2017

This is released as part of the Datadog Agent. See Datadog Agent ChangeLog for info about the other components of the agent.

Improvements

• Respect proxy_* configuration settings in datadog.conf. This allows the trace-agent to report to the Datadog backend through an http/s proxy in a similar manner to the infrastructure agent. #251

[dd-agent-release] 5.12.0 / 03-22-2017

This is released as part of the Datadog Agent. See Datadog Agent ChangeLog for info about the other components of the agent.

Improvements

• Watchdog restarts trace agent if it consumes too much memory or opens too many network connections. Default limits are very high for now (1 Gb memory, 5000 connections) so by default it has no impact. #182
• Reduce sampler log verbosity and expose internal state with Go expvar. #248
• Adaptive sampling. Sampling rate is updated in real-time, taking trace score into account. #228
• Weight spans using their sample rate. This can affect stats when sampling rare traces, as graphs are scaled up according to sampling rate. #226
• Stream decoding. This reduces the overall memory footprint of the agent (about 50% gain on a average) #227
• Quantize SQL AS aliases. A request such as SELECT server_table.host AS host_id FROM table would have SELECT server_table.host FROM TABLE as a resource. #225
• Updated the output of trace-agent info. #219

Bugfixes

• Now shells out to dd-agent to retrieve hostname. Prior to this change, the trace-agent would identify as a different host than the infra agent in subtle cases (EC2 / GCE for example). This created phantom hosts and general inconsistency in the UI #242
• Normalize environment strings in spans and in the configuration #233
• Fixed race issue between concentrator and sampler. #229

[dd-agent release] 5.11.0

This is the first release of datadog-trace-agent to be bundled into the datadog-agent package. The dd-trace-agent rpm and deb packages will be deprecated and will not receive further updates. This changelog will continue to be updated in tandem with future datadog-agent releases.

Improvements

• Use best-speed Gzip compression, improving agent encoding perf (4aac4f4)
• Standardize HTTP response statuses for the agent receiver (#213)
• Respect apm_enabled and non_local_traffic configuration flags. This is for consistency with the infrastructure agent, and ensures that running the trace-agent is an entirely opt-in process. (#200)
• Add an -info flag, allowing trace-agent vital statistics to be dumped to stdout via sudo service datadog-agent info(#214)
• Listen on port 8126 for traces, leaving port 7777 open for legacy clients. The port is meant to be contiguous with statsd 8125 , making it more straightforward to set firewall rules for example when statsd and trace-agent are both fielding requests. (#207)

Bugfixes

• Fix a bug in Redis quantization (#188)
• Fix signal handling on agent shutdown (#191)
• Add a WriteTimeout in the agent receiver to better handle long-running requests (7152b79)
• Fix a data race in the agent Sampler, during Add and Flush operations. This would occasionally cause us to deadlock and leak file descriptors. (#223)
• Fix a data race in connection leasing for the agent receiver. (#224)

For more details, see the full changeset

0.99.1101

Released versions

0.99.1101 for Debian/Ubuntu distributions
0.99.70 for RedHat/CentOS distributions

Improvements

• Use tinylib/msgp and typed msgpack decoding to improve serialization perf (#180)
• Improve lookup performance with a leaner struct for aggregating in-memory statistics from incoming traces. (#184)
• Improve memory allocation for truncated tags (#183)

Bugfixes

• Fix a bug in sampler logic that caused us to drop more traces than we should have (9ef64aa)
• Fix some statistic counts on sublayers (#185 #186)

For more details, see the full changeset