Ddtrace

This release introduces setting numeric tags as metrics, addresses a regression in Python 3 performance, and introduces a dual license.

Changes

Core

* Prefer random.getrandbits on Python 3+, fall back to urandom implementation on 2 (1183 -- thanks thieman)
* Set numeric tags on `Span.metrics` instead of `Span.meta` (1169, 1182)
* Update default sampler to new `DatadogSampler` (1172, 1166)
* Safely deprecate `ext` type constants (1165)

Tooling

* Fix botocore tests (1177)

Documentation

* Add Dual License (1181)
* Improve `tracer.trace` docs (1180 -- thanks adamchainz)
---

Changes

This patch release addresses an issue with installation:

* add funcsigs backport to install (1176)

Changeset

Read the [full changeset](https://github.com/DataDog/dd-trace-py/compare/v0.32.1...v0.32.2).
---

Changes

This patch release addresses an issue with installation:

* use environment markers for install requirements (1174 -- thanks JBKahn, 1175)

Changeset

Read the [full changeset](https://github.com/DataDog/dd-trace-py/compare/v0.32.0...v0.32.1).
---

This release adds support for dogpile.cache, fixes an issue with the bottle integration, addresses issues with CI, and makes several improvements to the core library.

Changes

Core

- Deprecating app type (1144, 1162)
- Refactor span types (1150)
- Use debtcollector (1152, 1147)
- Change DatadogSampler defaults (1151)
- Initial black formatting (1137, 1141)
- Health metrics improvements (1135, 1134, 1131)

Integrations

- Add support for dogpile.cache (1123 -- thanks goodspark)

Bug fixes

- Bottle: fix status code for error responses (1158)

Tooling

- Improve flake8 checks (1132)
- Pin multidict dependency for aiobotocore02 tests (1145 -- thanks codeboten)
- Fixing sqlalchemy test failures (1138 -- thanks codeboten)
- Remove unneeded tox dependencies (1124)
---

This release addresses issues with the gRPC, Celery, Elasticsearch integrations. In addition, there are internal improvements to how timing for spans is being handled.

Changes

Integrations

- celery: use strongrefs for celery signals (1122) fixes 1011
- elasticsearch: Add support for elasticsearch6 module (1089)
- grpc: improve handling exceptions (1117, 1119) and use callbacks to avoid waits (1097)
- opentracing: fix for compatibility tags (1096 -- thanks marshallbrekka)

Core and Internal

- core: replace time.time by monotonic clock (1109)
- core: rewrite agent writer on new process (1106)
- core: add support for dogstatsd unix socket (1101)
- core: always set rate limit metric (1060)
- internal: fix setting analytics sample rate of None (1120)
- internal: initial work on tracer health metrics (1130, 1129, 1127, 1125)
- internal: use args for LogRecord when logging (1116 -- thanks karolinepauls)
- span: use ns time (1113, 1112, 1105, 964)
- tracer: allow to override agent URL with a env var (1054)

Documentation

- docs: add a GitHub issue template (1118)
- Remove extra import from tracer get_call_context code snippet (1041)

Read the [full changeset](https://github.com/DataDog/dd-trace-py/compare/v0.30.2...v0.31.0) and the [release milestone](https://github.com/DataDog/dd-trace-py/milestone/47?closed=1).
---

Changes

This bug fix release introduces a fix for the agent writer that would cause traces to not be successfully flushed when the process forks.

Whenever we detect that we are in a new process (`os.getpid()` changes) we create a new queue to push finished traces into for flushing. In the previous version of the tracer (`v0.29.x`) we would also recreate the background thread that would read from this queue to ensure that we 1) had one background worker per-process and 2) had an updated reference to the new queue we created.

In the process of simplifying the code for the agent writer in `v0.30.0` we kept the code to recreate the queue, but didn't fully replicate the behavior for recreating the background thread. This meant that whenever the process was forked the background thread would maintain a reference to the old queue object and therefore be unable to pull traces from the queue we were pushing them into causing traces to get stuck in memory in the new queue (since no one was pulling from it).

With these changes we are recreating the entire background worker whenever the process is forked. This will cause us to once again create a new queue and new background thread for flushing finished traces.

Core and Internal

* core: rewrite agent writer on new process (1106)
* grpc: use callbacks to avoid waits (1097)

Changeset

Read the [full changeset](https://github.com/DataDog/dd-trace-py/compare/v0.30.1...v0.30.2).
---