datadog-checks-base 37.21.1__py2.py3-none-any.whl → 37.22.1__py2.py3-none-any.whl

datadog_checks/base/__about__.py

@@ -1,4 +1,4 @@
 # (C) Datadog, Inc. 2018-present
 # All rights reserved
 # Licensed under a 3-clause BSD style license (see LICENSE)
-__version__ = "37.21.1"
+__version__ = "37.22.1"

datadog_checks/base/utils/db/health.py

@@ -9,16 +9,20 @@ from __future__ import annotations
 import time
 from typing import TYPE_CHECKING
 
+from cachetools import TLRUCache
+
 from datadog_checks.base.utils.serialization import json
 
 if TYPE_CHECKING:
     from datadog_checks.base import DatabaseCheck
 try:
-    import datadog_agent
+    import datadog_agent  # type: ignore
 except ImportError:
     from datadog_checks.base.stubs import datadog_agent
 
 
+import threading
+import traceback
 from enum import Enum
 
 
@@ -28,6 +32,8 @@ class HealthEvent(Enum):
     """
 
     INITIALIZATION = 'initialization'
+    UNKNOWN_ERROR = 'unknown_error'
+    MISSED_COLLECTION = 'missed_collection'
 
 
 class HealthStatus(Enum):
@@ -40,6 +46,13 @@ class HealthStatus(Enum):
     ERROR = 'error'
 
 
+DEFAULT_COOLDOWN = 60 * 5
+
+
+def ttl(_key, value, now):
+    return now + value
+
+
 class Health:
     def __init__(self, check: DatabaseCheck):
         """
@@ -49,8 +62,18 @@ class Health:
             The check instance that will be used to submit health events.
         """
         self.check = check
-
-    def submit_health_event(self, name: HealthEvent, status: HealthStatus, tags: list[str] = None, **kwargs):
+        self._cache_lock = threading.Lock()
+        self._ttl_cache = TLRUCache(maxsize=1000, ttu=ttl)
+
+    def submit_health_event(
+        self,
+        name: HealthEvent,
+        status: HealthStatus,
+        tags: list[str] = None,
+        cooldown_time: int = None,
+        cooldown_values: list[str] = None,
+        data: dict = None,
+    ):
         """
         Submit a health event to the aggregator.
 
@@ -60,22 +83,52 @@ class Health:
             The health status to submit.
         :param tags: list of str
             Tags to associate with the health event.
-        :param kwargs: Additional keyword arguments to include in the event under `data`.
+        :param cooldown_time: int
+            The cooldown period in seconds to prevent the events with the same name and status
+            from being submitted again. If None there is no cooldown.
+        :param cooldown_values: list of str
+            Additional values to include in the cooldown key.
+        :param data: A dictionary to be submitted as `data`. Must be JSON serializable.
         """
+        category = self.check.__NAMESPACE__ or self.check.__class__.__name__.lower()
+        if cooldown_time:
+            cooldown_key = "|".join([category, name.value, status.value])
+            if cooldown_values:
+                cooldown_key = "|".join([cooldown_key, "|".join([f"{v}" for v in cooldown_values])])
+            with self._cache_lock:
+                if self._ttl_cache.get(cooldown_key, None):
+                    return
+                self._ttl_cache[cooldown_key] = cooldown_time
         self.check.event_platform_event(
             json.dumps(
                 {
                     'timestamp': time.time() * 1000,
                     'version': 1,
                     'check_id': self.check.check_id,
-                    'category': self.check.__NAMESPACE__ or self.check.__class__.__name__.lower(),
+                    'category': category,
                     'name': name,
                     'status': status,
                     'tags': tags or [],
                     'ddagentversion': datadog_agent.get_version(),
                     'ddagenthostname': datadog_agent.get_hostname(),
-                    'data': {**kwargs},
+                    'data': data,
                 }
             ),
             "dbm-health",
         )
+
+    def submit_exception_health_event(self, exception: Exception, data: dict):
+        trace = traceback.extract_tb(exception.__traceback__)
+        exc = trace.pop()
+        if exc:
+            self.submit_health_event(
+                name=HealthEvent.UNKNOWN_ERROR,
+                status=HealthStatus.ERROR,
+                data={
+                    "file": exc.filename,
+                    "line": exc.lineno,
+                    "function": exc.name,
+                    "exception_type": type(exception).__name__,
+                    **(data or {}),
+                },
+            )

datadog_checks/base/utils/db/utils.py

@@ -21,6 +21,7 @@ from datadog_checks.base import is_affirmative
 from datadog_checks.base.agent import datadog_agent
 from datadog_checks.base.log import get_check_logger
 from datadog_checks.base.utils.common import to_native_string
+from datadog_checks.base.utils.db.health import DEFAULT_COOLDOWN, HealthEvent, HealthStatus
 from datadog_checks.base.utils.db.types import Transformer  # noqa: F401
 from datadog_checks.base.utils.format import json
 from datadog_checks.base.utils.tracing import INTEGRATION_TRACING_SERVICE_NAME, tracing_enabled
@@ -109,7 +110,7 @@ class ConstantRateLimiter:
 
     def __init__(self, rate_limit_s, max_sleep_chunk_s=5):
         """
-        :param rate_limit_s: rate limit in seconds
+        :param rate_limit_s: rate limit in executions per second
         :param max_sleep_chunk_s: maximum size of each sleep chunk while waiting for the next period
         """
         self.rate_limit_s = max(rate_limit_s, 0)
@@ -293,10 +294,20 @@ class DBMAsyncJob(object):
         expected_db_exceptions=(),
         shutdown_callback=None,
         job_name=None,
+        # Some users may want to disable the missed collection event,
+        # for example if they set the collection interval intentionally low
+        # to effectively run the job in a loop
+        enable_missed_collection_event=True,
+        # List of features depenedent on the job running
+        # Defaults to [None] during init so that if no features are specified there will
+        # still be health events submitted for the job
+        features=None,
     ):
         self._check = check
         self._config_host = config_host
+        # The min_collection_interval is the expected collection interval for the main check
         self._min_collection_interval = min_collection_interval
+        self._expected_collection_interval = 1 / rate_limit if rate_limit > 0 else 0
         # map[dbname -> psycopg connection]
         self._log = get_check_logger()
         self._job_loop_future = None
@@ -314,6 +325,10 @@ class DBMAsyncJob(object):
         self._enabled = enabled
         self._expected_db_exceptions = expected_db_exceptions
         self._job_name = job_name
+        self._enable_missed_collection_event = enable_missed_collection_event
+        self._features = features
+        if self._features is None:
+            self._features = [None]
 
     def cancel(self):
         """
@@ -342,6 +357,39 @@ class DBMAsyncJob(object):
         elif self._job_loop_future is None or not self._job_loop_future.running():
             self._job_loop_future = DBMAsyncJob.executor.submit(self._job_loop)
         else:
+            if (
+                hasattr(self._check, 'health')
+                and self._enable_missed_collection_event
+                and self._expected_collection_interval >= 1
+                and self._last_run_start
+            ):
+                # Assume a collection interval of less than 1 second is an attempt to run the job in a loop
+                elapsed_time = time.time() - self._last_run_start
+                if elapsed_time > self._expected_collection_interval:
+                    # Missed a collection interval, submit a health event for each feature that depends on this job
+                    for feature in self._features:
+                        self._check.health.submit_health_event(
+                            name=HealthEvent.MISSED_COLLECTION,
+                            status=HealthStatus.WARNING,
+                            tags=self._job_tags,
+                            # Use a cooldown to avoid spamming if the job is missing the collection interval
+                            # in a flappy manner
+                            cooldown_time=DEFAULT_COOLDOWN,
+                            cooldown_values=[self._dbms, self._job_name],
+                            data={
+                                "dbms": self._dbms,
+                                "job_name": self._job_name,
+                                # send in ms for consistency with other metrics
+                                "last_run_start": self._last_run_start * 1000,
+                                "elapsed_time": elapsed_time * 1000,
+                                "expected_collection_interval": self._expected_collection_interval * 1000,
+                                "feature": feature,
+                            },
+                        )
+                    self._check.count(
+                        "dd.{}.async_job.missed_collection".format(self._dbms), 1, tags=self._job_tags, raw=True
+                    )
+
             self._log.debug("Job loop already running. job=%s", self._job_name)
 
     def _job_loop(self):
@@ -394,6 +442,14 @@ class DBMAsyncJob(object):
                     tags=self._job_tags + ["error:crash-{}".format(type(e))],
                     raw=True,
                 )
+
+                if hasattr(self._check, 'health'):
+                    try:
+                        self._check.health.submit_exception_health_event(e, data={"job_name": self._job_name})
+                    except Exception as health_error:
+                        self._log.exception(
+                            "[%s] Failed to submit error health event", self._job_tags_str, health_error
+                        )
         finally:
             self._log.info("[%s] Shutting down job loop", self._job_tags_str)
             if self._shutdown_callback:
@@ -410,6 +466,7 @@ class DBMAsyncJob(object):
 
     def _run_job_rate_limited(self):
         try:
+            self._last_run_start = time.time()
             self._run_job_traced()
         except:
             raise

datadog_checks/base/utils/tls.py

@@ -56,6 +56,67 @@ class TlsConfig(BaseModel, frozen=True):
     tls_verify: bool = True
 
 
+def _load_certifi_fallback(context):
+    """Attempt to load CA certificates from certifi as a fallback."""
+    try:
+        import certifi
+    except ImportError:
+        LOGGER.warning('Failed to import certifi, TLS verification may fail.')
+        return
+
+    try:
+        certifi_path = certifi.where()
+        context.load_verify_locations(cafile=certifi_path)
+        LOGGER.info('Successfully loaded CA certificates from certifi bundle: %s', certifi_path)
+    except (FileNotFoundError, IOError) as e:
+        LOGGER.error('Failed to load CA certificates from certifi bundle: %s. TLS verification may fail.', e)
+    except Exception as e:
+        LOGGER.error('Unexpected error loading certifi certificates: %s', e)
+
+
+def _load_ca_certs(context, config):
+    # https://docs.python.org/3/library/ssl.html#ssl.SSLContext.load_verify_locations
+    # https://docs.python.org/3/library/ssl.html#ssl.SSLContext.load_default_certs
+    ca_cert = config.get('tls_ca_cert')
+
+    # Handle user-provided CA cert
+    if ca_cert:
+        ca_cert = os.path.expanduser(ca_cert)
+        try:
+            if os.path.isdir(ca_cert):
+                context.load_verify_locations(cafile=None, capath=ca_cert, cadata=None)
+            else:
+                context.load_verify_locations(cafile=ca_cert, capath=None, cadata=None)
+        except FileNotFoundError:
+            LOGGER.warning(
+                'TLS CA certificate file not found: %s. Please check the `tls_ca_cert` configuration option.',
+                ca_cert,
+            )
+            return
+    else:
+        # Try to load system default certs
+        try:
+            context.load_default_certs(ssl.Purpose.SERVER_AUTH)
+        except Exception as e:
+            LOGGER.debug('Failed to load default CA certificates: %s', e)
+
+        # Check if any certs were actually loaded
+        if not context.get_ca_certs():
+            LOGGER.info('No CA certificates loaded from system default paths, attempting certifi fallback.')
+            _load_certifi_fallback(context)
+
+    # Load intermediate CA certs if provided
+    intermediate_ca_certs = config.get('tls_intermediate_ca_certs')
+    if intermediate_ca_certs:
+        try:
+            context.load_verify_locations(cadata='\n'.join(intermediate_ca_certs))
+        except ssl.SSLError:
+            LOGGER.warning(
+                "TLS intermediate CA certificate(s) could not be loaded: %s. ",
+                intermediate_ca_certs,
+            )
+
+
 def create_ssl_context(config):
     # https://docs.python.org/3/library/ssl.html#ssl.SSLContext
     # https://docs.python.org/3/library/ssl.html#ssl.PROTOCOL_TLS_CLIENT
@@ -78,32 +139,10 @@ def create_ssl_context(config):
         LOGGER.debug('Setting TLS ciphers to: %s', configured_ciphers)
         context.set_ciphers(configured_ciphers)
 
-    # https://docs.python.org/3/library/ssl.html#ssl.SSLContext.load_verify_locations
-    # https://docs.python.org/3/library/ssl.html#ssl.SSLContext.load_default_certs
-    ca_cert = config.get('tls_ca_cert')
-    try:
-        if ca_cert:
-            ca_cert = os.path.expanduser(ca_cert)
-            if os.path.isdir(ca_cert):
-                context.load_verify_locations(cafile=None, capath=ca_cert, cadata=None)
-            else:
-                context.load_verify_locations(cafile=ca_cert, capath=None, cadata=None)
-        else:
-            context.load_default_certs(ssl.Purpose.SERVER_AUTH)
-    except FileNotFoundError:
-        LOGGER.warning(
-            'TLS CA certificate file not found: %s. Please check the `tls_ca_cert` configuration option.',
-            ca_cert,
-        )
-    intermediate_ca_certs = config.get('tls_intermediate_ca_certs')
-    try:
-        if intermediate_ca_certs:
-            context.load_verify_locations(cadata='\n'.join(intermediate_ca_certs))
-    except ssl.SSLError:
-        LOGGER.warning(
-            "TLS intermediate CA certificate(s) could not be loaded: %s. ",
-            intermediate_ca_certs,
-        )
+    if context.verify_mode == ssl.CERT_NONE:
+        LOGGER.debug('TLS verification is disabled; skipping CA certificate configuration.')
+    else:
+        _load_ca_certs(context, config)
 
     # https://docs.python.org/3/library/ssl.html#ssl.SSLContext.load_cert_chain
     client_cert, client_key = config.get('tls_cert'), config.get('tls_private_key')

{datadog_checks_base-37.21.1.dist-info → datadog_checks_base-37.22.1.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: datadog-checks-base
-Version: 37.21.1
+Version: 37.22.1
 Summary: The Datadog Check Toolkit
 Project-URL: Source, https://github.com/DataDog/integrations-core
 Author-email: Datadog <packages@datadoghq.com>

{datadog_checks_base-37.21.1.dist-info → datadog_checks_base-37.22.1.dist-info}/RECORD

@@ -3,7 +3,7 @@ datadog_checks/config.py,sha256=PrAXGdlLnoV2VMQff_noSaSJJ0wg4BAiGnw7jCQLSik,196
 datadog_checks/errors.py,sha256=eFwmnrX-batIgbu-iJyseqAPNO_4rk1UuaKK89evLhg,155
 datadog_checks/log.py,sha256=orvOgMKGNEsqSTLalCAQpWP-ouorpG1A7Gn-j2mRD80,301
 datadog_checks/py.typed,sha256=la67KBlbjXN-_-DfGNcdOcjYumVpKG_Tkw-8n5dnGB4,8
-datadog_checks/base/__about__.py,sha256=zY9yrvpmsQJIlSMC9LByfIb_KGfdfnD6iYbUnIQNZBo,139
+datadog_checks/base/__about__.py,sha256=LHTbj6b1TjdvOrdvo82IsglGOmoTwkDhLPngNctW7FA,139
 datadog_checks/base/__init__.py,sha256=yWegSLE-TZWIGSvAiJj9PSrUxzlOo_UVJLt2zORZ8Ek,363
 datadog_checks/base/__init__.pyi,sha256=a4Y1JIcPJ8pz9tRkBAvjWdtvSQwZxbMZBuRmIiSs_4E,1031
 datadog_checks/base/agent.py,sha256=nX9x_BYYizRKGNYfXq5z7S0FZ9xcX_wd2tuxpGe3_8k,350
@@ -127,7 +127,7 @@ datadog_checks/base/utils/tagging.py,sha256=a-0_pXGOwg4ewmDiJon7rM_96ZJvVpHVMwhy
 datadog_checks/base/utils/tailfile.py,sha256=QZmH8eCxeClFWRMZn6zDwgOx1y5dAXc5yI_uzECj0ak,3892
 datadog_checks/base/utils/time.py,sha256=cNy7CtsJzSUMi7J-3WReZVUvNyYOGkJwItqJMY01qqA,1373
 datadog_checks/base/utils/timeout.py,sha256=eOBZofFN-hOu5xJeMOF3ac_ofcy9EY6-kvmBvP4QEFg,2140
-datadog_checks/base/utils/tls.py,sha256=DoJqqpnWTC8up_UF4AL_UUBJ9Sb-1Fvwjd2gSxmiVBY,7434
+datadog_checks/base/utils/tls.py,sha256=to_33MaSEjvo6zvqRm9gKRx_Zk9Ayq53tnZiuKvNy24,8922
 datadog_checks/base/utils/tracing.py,sha256=2QZjQKNKvowRYFpqOQc4sUoYUWu6JJKJofk-rvstUgY,6076
 datadog_checks/base/utils/tracking.py,sha256=FYIouqu3KB-JsxgxM1iX5Ipv_cWhs8zQOTN4SxXOkJ4,3701
 datadog_checks/base/utils/agent/__init__.py,sha256=o3aWvy3PhykD_h7YT3s628O0W2YpHis0NlQsSV1PI04,115
@@ -141,7 +141,7 @@ datadog_checks/base/utils/concurrency/limiter.py,sha256=is2ZpUEjfsI4nBGtXG2D0Zgv
 datadog_checks/base/utils/db/__init__.py,sha256=EVTc2FtnHWLHXI3M79jyMn9ypZAMa9eqG3EKLAiMF-M,211
 datadog_checks/base/utils/db/__init__.pyi,sha256=ewmGxxyJ52wAaYxNZahi2koEUnddfvHcn3HYxQ3RUr0,240
 datadog_checks/base/utils/db/core.py,sha256=bYanwXIqBzsSxK7b-Ofb0W1WiHbFBtKyYdUBonBLe_Q,11165
-datadog_checks/base/utils/db/health.py,sha256=rdcZPdlLMT4g9tepkietuViKtmP4gwPsjMGCvqC5s54,2366
+datadog_checks/base/utils/db/health.py,sha256=riJaJInOuYFK3y0wNH19HojIkXPwQaMtvxuyPCRxnZY,4182
 datadog_checks/base/utils/db/query.py,sha256=-PyxdqpbShkQ78h7sWnghQZVtjdLGVrm71n8OpHuPW4,14432
 datadog_checks/base/utils/db/sql.py,sha256=oiEzQa_vC_w3U65VFrFCoQHWj5GQLLRlSO0CfiSlp4A,2490
 datadog_checks/base/utils/db/sql_commenter.py,sha256=r_efK6TGRQxM_-Qj-ndEJdECk47J4nCFjkVyxu1XmvU,1522
@@ -149,7 +149,7 @@ datadog_checks/base/utils/db/statement_metrics.py,sha256=U7EtERkmFzfCtfyd3094fBa
 datadog_checks/base/utils/db/timed_cache.py,sha256=a9Ks5KKUvExB6GOATXTSCLamVtLD919Dn6HpweGKtFw,2114
 datadog_checks/base/utils/db/transform.py,sha256=H3JN8_MF0Pk0HaXvIZeX1A8iQrP8KBgS741MPeBiWDo,23969
 datadog_checks/base/utils/db/types.py,sha256=OLX2Oq58JQPFBD4oqUpCLkAP7ovRGN_i1vFk1E0N8Lg,267
-datadog_checks/base/utils/db/utils.py,sha256=g6GAG1OGvo4O-GrTB6nETvUZGfVUSpqYYJMBuhLs2pM,23130
+datadog_checks/base/utils/db/utils.py,sha256=fJZrf2YcAbWmX4IRqLf4r0Rt8tGLDwgdIVmSMSft1co,26464
 datadog_checks/base/utils/discovery/__init__.py,sha256=vPCOdsThBcBjFJRPhDm6IsZGOwk8HlvciwCe_l8dKLk,211
 datadog_checks/base/utils/discovery/__init__.pyi,sha256=ScVLU1Njj9ekZmewltb0cULI6BylssVHfn4CcPNeyr8,173
 datadog_checks/base/utils/discovery/cache.py,sha256=f9L3A7YZpZ-mpZpFIwjsa5ab9cZMGkqdetdr9EpalbI,887
@@ -223,6 +223,6 @@ datadog_checks/utils/tracing.py,sha256=HQbQakKM-Lw75MDkItaYJYipS6YO24Z_ymDVxDsx5
 datadog_checks/utils/prometheus/__init__.py,sha256=8WwXnM9g1sfS5267QYCJX_hd8MZl5kRgBgQ_SzdNdXs,161
 datadog_checks/utils/prometheus/functions.py,sha256=4vWsTGLgujHwdYZo0tlAQkqDPHofqUJM3k9eItJqERQ,197
 datadog_checks/utils/prometheus/metrics_pb2.py,sha256=xg3UdUHe4TjeR4s13LUKZ2U1WVSt6U6zjsVRG6lX6dc,173
-datadog_checks_base-37.21.1.dist-info/METADATA,sha256=W1zsqHGVigXb-0R8vJOUv2kxuRDn0S02dYFWEfggkHQ,4245
-datadog_checks_base-37.21.1.dist-info/WHEEL,sha256=tkmg4JIqwd9H8mL30xA7crRmoStyCtGp0VWshokd1Jc,105
-datadog_checks_base-37.21.1.dist-info/RECORD,,
+datadog_checks_base-37.22.1.dist-info/METADATA,sha256=93hddxC1RfHZzTr2DIbxWE5_78AWLYmqasrpxbej7ng,4245
+datadog_checks_base-37.22.1.dist-info/WHEEL,sha256=tkmg4JIqwd9H8mL30xA7crRmoStyCtGp0VWshokd1Jc,105
+datadog_checks_base-37.22.1.dist-info/RECORD,,