Merge branch 'main' into brettlangdon/simplify.flask_simple

.gitlab-ci.yml

@@ -3,10 +3,12 @@ stages:
   - deploy
   - benchmarks
   - benchmarks-pr-comment
+  - macrobenchmarks
 
 include:
   - remote: https://gitlab-templates.ddbuild.io/apm/packaging.yml
   - local: ".gitlab/benchmarks.yml"
+  - local: ".gitlab/macrobenchmarks.yml"
 
 variables:
   DOWNSTREAM_BRANCH:

.gitlab/macrobenchmarks.yml

@@ -0,0 +1,86 @@
+variables:
+  BASE_CI_IMAGE: 486234852809.dkr.ecr.us-east-1.amazonaws.com/ci/benchmarking-platform:dd-trace-py-macrobenchmarks
+
+.macrobenchmarks:
+  stage: macrobenchmarks
+  needs: []
+  tags: ["runner:apm-k8s-same-cpu"]
+  timeout: 1h
+  rules:
+    - if: $CI_PIPELINE_SOURCE == "schedule"
+      when: always
+    - when: manual
+    ## Next step, enable:
+    # - if: $CI_COMMIT_REF_NAME == "main"
+    #   when: always
+  # If you have a problem with Gitlab cache, see Troubleshooting section in Benchmarking Platform docs
+  image: $BENCHMARKS_CI_IMAGE
+  script: |
+    git clone --branch python/macrobenchmarks https://gitlab-ci-token:${CI_JOB_TOKEN}@gitlab.ddbuild.io/DataDog/benchmarking-platform platform && cd platform
+    if [ "$BP_PYTHON_SCENARIO_DIR" == "flask-realworld" ]; then
+      bp-runner bp-runner.flask-realworld.yml --debug
+    else
+      bp-runner bp-runner.simple.yml --debug
+    fi
+  artifacts:
+    name: "artifacts"
+    when: always
+    paths:
+      - platform/artifacts/
+    expire_in: 3 months
+  variables:
+    # Benchmark's env variables. Modify to tweak benchmark parameters.
+    DD_TRACE_DEBUG: "false"
+    DD_RUNTIME_METRICS_ENABLED: "true"
+    DD_REMOTE_CONFIGURATION_ENABLED: "false"
+    DD_INSTRUMENTATION_TELEMETRY_ENABLED: "false"
+
+    K6_OPTIONS_NORMAL_OPERATION_RATE: 40
+    K6_OPTIONS_NORMAL_OPERATION_DURATION: 5m
+    K6_OPTIONS_NORMAL_OPERATION_GRACEFUL_STOP: 1m
+    K6_OPTIONS_NORMAL_OPERATION_PRE_ALLOCATED_VUS: 4
+    K6_OPTIONS_NORMAL_OPERATION_MAX_VUS: 4
+
+    K6_OPTIONS_HIGH_LOAD_RATE: 500
+    K6_OPTIONS_HIGH_LOAD_DURATION: 1m
+    K6_OPTIONS_HIGH_LOAD_GRACEFUL_STOP: 30s
+    K6_OPTIONS_HIGH_LOAD_PRE_ALLOCATED_VUS: 4
+    K6_OPTIONS_HIGH_LOAD_MAX_VUS: 4
+
+    # Gitlab and BP specific env vars. Do not modify.
+    FF_USE_LEGACY_KUBERNETES_EXECUTION_STRATEGY: "true"
+
+  # Workaround: Currently we're not running the benchmarks on every PR, but GitHub still shows them as pending.
+  # By marking the benchmarks as allow_failure, this should go away. (This workaround should be removed once the
+  # benchmarks get changed to run on every PR)
+  allow_failure: true
+
+macrobenchmarks:
+  extends: .macrobenchmarks
+  parallel:
+    matrix:
+      - DD_BENCHMARKS_CONFIGURATION: baseline
+        BP_PYTHON_SCENARIO_DIR: flask-realworld
+        DDTRACE_INSTALL_VERSION: "git+https://github.com/Datadog/dd-trace-py@${CI_COMMIT_SHA}"
+
+      - DD_BENCHMARKS_CONFIGURATION: only-tracing
+        BP_PYTHON_SCENARIO_DIR: flask-realworld
+        DDTRACE_INSTALL_VERSION: "git+https://github.com/Datadog/dd-trace-py@${CI_COMMIT_SHA}"
+
+      - DD_BENCHMARKS_CONFIGURATION: only-tracing
+        BP_PYTHON_SCENARIO_DIR: flask-realworld
+        DDTRACE_INSTALL_VERSION: "git+https://github.com/Datadog/dd-trace-py@${CI_COMMIT_SHA}"
+        DD_REMOTE_CONFIGURATION_ENABLED: "false"
+        DD_INSTRUMENTATION_TELEMETRY_ENABLED: "true"
+
+      - DD_BENCHMARKS_CONFIGURATION: only-tracing
+        BP_PYTHON_SCENARIO_DIR: flask-realworld
+        DDTRACE_INSTALL_VERSION: "git+https://github.com/Datadog/dd-trace-py@${CI_COMMIT_SHA}"
+        DD_REMOTE_CONFIGURATION_ENABLED: "false"
+        DD_INSTRUMENTATION_TELEMETRY_ENABLED: "false"
+
+      - DD_BENCHMARKS_CONFIGURATION: only-tracing
+        BP_PYTHON_SCENARIO_DIR: flask-realworld
+        DDTRACE_INSTALL_VERSION: "git+https://github.com/Datadog/dd-trace-py@${CI_COMMIT_SHA}"
+        DD_REMOTE_CONFIGURATION_ENABLED: "true"
+        DD_INSTRUMENTATION_TELEMETRY_ENABLED: "true"

ddtrace/_trace/tracer.py

@@ -230,7 +230,8 @@ def __init__(
 
         self.enabled = config._tracing_enabled
         self.context_provider = context_provider or DefaultContextProvider()
-        self._user_sampler: Optional[BaseSampler] = None
+        # _user_sampler is the backup in case we need to revert from remote config to local
+        self._user_sampler: Optional[BaseSampler] = DatadogSampler()
         self._sampler: BaseSampler = DatadogSampler()
         self._dogstatsd_url = agent.get_stats_url() if dogstatsd_url is None else dogstatsd_url
         self._compute_stats = config._trace_compute_stats
@@ -286,7 +287,7 @@ def __init__(
         self._shutdown_lock = RLock()
 
         self._new_process = False
-        config._subscribe(["_trace_sample_rate"], self._on_global_config_update)
+        config._subscribe(["_trace_sample_rate", "_trace_sampling_rules"], self._on_global_config_update)
         config._subscribe(["logs_injection"], self._on_global_config_update)
         config._subscribe(["tags"], self._on_global_config_update)
         config._subscribe(["_tracing_enabled"], self._on_global_config_update)
@@ -1125,19 +1126,10 @@ def _is_span_internal(span):
 
     def _on_global_config_update(self, cfg, items):
         # type: (Config, List) -> None
-        if "_trace_sample_rate" in items:
-            # Reset the user sampler if one exists
-            if cfg._get_source("_trace_sample_rate") != "remote_config" and self._user_sampler:
-                self._sampler = self._user_sampler
-                return
-
-            if cfg._get_source("_trace_sample_rate") != "default":
-                sample_rate = cfg._trace_sample_rate
-            else:
-                sample_rate = None
 
-            sampler = DatadogSampler(default_sample_rate=sample_rate)
-            self._sampler = sampler
+        # sampling configs always come as a pair
+        if "_trace_sample_rate" in items and "_trace_sampling_rules" in items:
+            self._handle_sampler_update(cfg)
 
         if "tags" in items:
             self._tags = cfg.tags.copy()
@@ -1160,3 +1152,42 @@ def _on_global_config_update(self, cfg, items):
                 from ddtrace.contrib.logging import unpatch
 
                 unpatch()
+
+    def _handle_sampler_update(self, cfg):
+        # type: (Config) -> None
+        if (
+            cfg._get_source("_trace_sample_rate") != "remote_config"
+            and cfg._get_source("_trace_sampling_rules") != "remote_config"
+            and self._user_sampler
+        ):
+            # if we get empty configs from rc for both sample rate and rules, we should revert to the user sampler
+            self.sampler = self._user_sampler
+            return
+
+        if cfg._get_source("_trace_sample_rate") != "remote_config" and self._user_sampler:
+            try:
+                sample_rate = self._user_sampler.default_sample_rate  # type: ignore[attr-defined]
+            except AttributeError:
+                log.debug("Custom non-DatadogSampler is being used, cannot pull default sample rate")
+                sample_rate = None
+        elif cfg._get_source("_trace_sample_rate") != "default":
+            sample_rate = cfg._trace_sample_rate
+        else:
+            sample_rate = None
+
+        if cfg._get_source("_trace_sampling_rules") != "remote_config" and self._user_sampler:
+            try:
+                sampling_rules = self._user_sampler.rules  # type: ignore[attr-defined]
+                # we need to chop off the default_sample_rate rule so the new sample_rate can be applied
+                sampling_rules = sampling_rules[:-1]
+            except AttributeError:
+                log.debug("Custom non-DatadogSampler is being used, cannot pull sampling rules")
+                sampling_rules = None
+        elif cfg._get_source("_trace_sampling_rules") != "default":
+            sampling_rules = DatadogSampler._parse_rules_from_str(cfg._trace_sampling_rules)
+        else:
+            sampling_rules = None
+
+        sampler = DatadogSampler(rules=sampling_rules, default_sample_rate=sample_rate)
+
+        self._sampler = sampler

ddtrace/internal/constants.py

@@ -90,7 +90,9 @@
 
 class _PRIORITY_CATEGORY:
     USER = "user"
-    RULE = "rule"
+    RULE_DEF = "rule_default"
+    RULE_CUSTOMER = "rule_customer"
+    RULE_DYNAMIC = "rule_dynamic"
     AUTO = "auto"
     DEFAULT = "default"
 
@@ -99,7 +101,9 @@ class _PRIORITY_CATEGORY:
 # used to simplify code that selects sampling priority based on many factors
 _CATEGORY_TO_PRIORITIES = {
     _PRIORITY_CATEGORY.USER: (USER_KEEP, USER_REJECT),
-    _PRIORITY_CATEGORY.RULE: (USER_KEEP, USER_REJECT),
+    _PRIORITY_CATEGORY.RULE_DEF: (USER_KEEP, USER_REJECT),
+    _PRIORITY_CATEGORY.RULE_CUSTOMER: (USER_KEEP, USER_REJECT),
+    _PRIORITY_CATEGORY.RULE_DYNAMIC: (USER_KEEP, USER_REJECT),
     _PRIORITY_CATEGORY.AUTO: (AUTO_KEEP, AUTO_REJECT),
     _PRIORITY_CATEGORY.DEFAULT: (AUTO_KEEP, AUTO_REJECT),
 }