feat(config): Align Sentry events with the backend

packages/dsw-config/CHANGELOG.md

@@ -8,9 +8,15 @@ to [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
 ## [Unreleased]
 
 ### Added
+- `SentryReporter.set_command` / `reset_command` tag events with the persistent command being processed the way the backend does: `function`, `tenantUuid` and `traceUuid` tags, the creator as the user, the command UUID as the trace ID, and `uuid`, `attempts` and `maxAttempts` extras
 
 - `read_config` loads the configuration YAML from AWS AppConfig when `AWS_APP_CONFIG` is set (environment `Default`, profile derived from the file name, overridable by `AWS_APP_CONFIG_ENVIRONMENT` / `AWS_APP_CONFIG_PROFILE`), and from the local file otherwise; a warm Lambda reuses its session and re-polls only after the poll interval
 
+### Changed
+
+- `sentry.environment` (`SENTRY_ENVIRONMENT`) defaults to `development` instead of `production`, as in the backend, so a deployment that does not set it no longer reports as production
+- `SentryReporter.initialize` takes `component` (the `component` tag) instead of `prog_name`
+
 ## [4.34.0]
 
 ### Added

packages/dsw-config/dsw/config/keys.py

@@ -202,7 +202,7 @@ class _SentryKeys(ConfigKeysContainer):
     environment = ConfigKey(
         yaml_path=['sentry', 'environment'],
         var_names=['SENTRY_ENVIRONMENT'],
-        default='production',
+        default='development',
         cast=cast_str,
     )
 

packages/dsw-config/dsw/config/sentry.py

@@ -2,6 +2,7 @@
 
 import logging
 import typing
+import uuid
 
 import sentry_sdk
 from sentry_sdk.integrations.logging import LoggingIntegration
@@ -18,12 +19,21 @@
 LOG = logging.getLogger(__name__)
 
 
+class Command(typing.Protocol):
+    uuid: str
+    function: str
+    tenant_uuid: str
+    created_by: str | None
+    attempts: int
+    max_attempts: int
+
+
 class SentryReporter:
     report = False
     filters: list[EventProcessor] = []
 
     @classmethod
-    def initialize(cls, *, config: SentryConfig, prog_name: str, release: str,
+    def initialize(cls, *, config: SentryConfig, component: str, release: str,
                    breadcrumb_level: int | None = logging.INFO,
                    event_level: int | None = logging.ERROR):
         if config.enabled and not config.workers_dsn:
@@ -52,7 +62,7 @@ def before_send(event, hint):
                     ),
                 ],
             )
-            sentry_sdk.set_tag('component', prog_name)
+            sentry_sdk.set_tag('component', component)
 
     @classmethod
     def capture_exception(cls, *args, **kwargs):
@@ -68,3 +78,43 @@ def capture_message(cls, *args, **kwargs):
     def set_tags(cls, **tags):
         if cls.report:
             sentry_sdk.set_tags(tags)
+
+    @classmethod
+    def set_command(cls, command: Command, *, tenant_uuid: str | None = None):
+        """Tag events with the persistent command being processed, the way the backend does.
+
+        The command UUID is the trace UUID of the job (it is the `traceId` in the logs).
+        """
+        if not cls.report:
+            return
+        sentry_sdk.set_tags({
+            'function': command.function,
+            'tenantUuid': tenant_uuid or command.tenant_uuid,
+            'traceUuid': command.uuid,
+        })
+        sentry_sdk.set_extra('uuid', command.uuid)
+        sentry_sdk.set_extra('attempts', str(command.attempts))
+        sentry_sdk.set_extra('maxAttempts', str(command.max_attempts))
+        sentry_sdk.set_user({'id': command.created_by} if command.created_by else None)
+        trace_id = _to_trace_id(command.uuid)
+        if trace_id is not None:
+            sentry_sdk.continue_trace({'sentry-trace': f'{trace_id}-{trace_id[:16]}'})
+
+    @classmethod
+    def reset_command(cls):
+        if not cls.report:
+            return
+        scope = sentry_sdk.get_isolation_scope()
+        for tag in ('function', 'tenantUuid', 'traceUuid'):
+            scope.remove_tag(tag)
+        for extra in ('uuid', 'attempts', 'maxAttempts'):
+            scope.remove_extra(extra)
+        scope.set_user(None)
+        scope.set_new_propagation_context()
+
+
+def _to_trace_id(value: str) -> str | None:
+    try:
+        return uuid.UUID(value).hex
+    except ValueError:
+        return None

packages/dsw-config/tests/test_parser.py

@@ -91,3 +91,7 @@ def test_sentry_defaults():
     assert sentry.workers_dsn is None
     assert sentry.max_breadcrumbs is None
     assert sentry.traces_sample_rate is None
+
+
+def test_sentry_environment_defaults_to_development():
+    assert make_parser().sentry.environment == 'development'

packages/dsw-config/tests/test_sentry.py

@@ -1,6 +1,8 @@
+import dataclasses
 import logging
 
 import pytest
+import sentry_sdk
 
 from dsw.config.model import SentryConfig
 from dsw.config.sentry import SentryReporter
@@ -25,7 +27,7 @@ def _reset_reporter():
 def test_not_reporting_when_disabled():
     SentryReporter.initialize(
         config=make_config(enabled=False, dsn='https://token@sentry.io/1'),
-        prog_name='test',
+        component='test',
         release='0.0.0',
     )
     assert SentryReporter.report is False
@@ -36,8 +38,75 @@ def test_not_reporting_without_dsn(caplog, dsn):
     with caplog.at_level(logging.WARNING):
         SentryReporter.initialize(
             config=make_config(enabled=True, dsn=dsn),
-            prog_name='test',
+            component='test',
             release='0.0.0',
         )
     assert SentryReporter.report is False
     assert 'no DSN' in caplog.text
+
+
+@dataclasses.dataclass
+class FakeCommand:
+    uuid: str = '8a7d1f0e-3b2c-4d5e-9f60-7182a3b4c5d6'
+    function: str = 'generateDocument'
+    tenant_uuid: str = '00000000-0000-0000-0000-000000000000'
+    created_by: str | None = 'c1d2e3f4-a5b6-4c7d-8e9f-0a1b2c3d4e5f'
+    attempts: int = 1
+    max_attempts: int = 10
+
+
+@pytest.fixture
+def captured():
+    events = []
+
+    def capture(event, _hint):
+        events.append(event)
+        return None
+
+    SentryReporter.initialize(
+        config=make_config(enabled=True, dsn='https://token@sentry.example.org/1'),
+        component='document_worker',
+        release='v4.0.0~abc1234',
+    )
+    SentryReporter.filters.append(capture)
+    yield events
+    SentryReporter.filters.remove(capture)
+    sentry_sdk.get_isolation_scope().clear()
+    sentry_sdk.get_client().close()
+
+
+def test_command_tags_match_backend(captured):
+    command = FakeCommand()
+    SentryReporter.set_command(command)
+    SentryReporter.capture_message('failed')
+
+    event = captured[-1]
+    assert event['environment'] == 'test'
+    assert event['release'] == 'v4.0.0~abc1234'
+    assert event['tags'] == {
+        'component': 'document_worker',
+        'function': 'generateDocument',
+        'tenantUuid': command.tenant_uuid,
+        'traceUuid': command.uuid,
+    }
+    assert event['user'] == {'id': command.created_by}
+    assert event['extra']['maxAttempts'] == '10'
+    assert event['contexts']['trace']['trace_id'] == command.uuid.replace('-', '')
+
+
+def test_command_tenant_override(captured):
+    SentryReporter.set_command(FakeCommand(), tenant_uuid='t')
+    SentryReporter.capture_message('failed')
+    assert captured[-1]['tags']['tenantUuid'] == 't'
+
+
+def test_reset_command(captured):
+    command = FakeCommand()
+    SentryReporter.set_command(command)
+    SentryReporter.reset_command()
+    SentryReporter.capture_message('failed')
+
+    event = captured[-1]
+    assert event['tags'] == {'component': 'document_worker'}
+    assert 'user' not in event
+    assert event['contexts']['trace']['trace_id'] != command.uuid.replace('-', '')

packages/dsw-data-seeder/CHANGELOG.md

@@ -11,6 +11,10 @@ to [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
 
 - Lambda handler reads its configuration from AWS AppConfig when `AWS_APP_CONFIG` is set (see `dsw-config`)
 
+### Changed
+
+- Sentry events carry the same tags as the backend's: `component` is the persistent command component (`data_seeder`), and a command sets `function`, `tenantUuid`, `traceUuid` (the command UUID, the `traceId` of the logs), the user who created it, and its trace; the tenant is the one being seeded, `recipe_name` is `recipeName` and `command_uuid` / `tenant_uuid` are gone
+
 ## [4.34.0]
 
 ### Fixed

packages/dsw-data-seeder/dsw/data_seeder/seeder.py

@@ -323,7 +323,7 @@ def _init_sentry(self):
         SentryReporter.initialize(
             config=self.cfg.sentry,
             release=BUILD_INFO.version,
-            prog_name=consts.PROG_NAME,
+            component=consts.CMD_COMPONENT,
         )
 
     def _init_extra_connections(self):
@@ -339,7 +339,7 @@ def _init_extra_connections(self):
             )
 
     def _prepare_recipe(self, recipe_name: str):
-        SentryReporter.set_tags(recipe_name=recipe_name)
+        SentryReporter.set_tags(recipeName=recipe_name)
         LOG.info('Loading recipe "%s"', recipe_name)
         self.recipes = SeedRecipe.load_from_dir(self.workdir)
         if recipe_name not in self.recipes:
@@ -375,8 +375,8 @@ def run_once(self):
 
     def work(self, command: PersistentCommand):
         Context.get().update_trace_id(command.uuid)
-        SentryReporter.set_tags(command_uuid=command.uuid)
         tenant_uuid: str = command.body['tenantUuid']
+        SentryReporter.set_command(command, tenant_uuid=tenant_uuid)
         recipe_name: str | None = command.body.get('recipe')
         LOG.debug('Processing command: %s (tenant: %s, seed: %s)',
                   command.uuid, tenant_uuid, recipe_name)
@@ -390,7 +390,7 @@ def work(self, command: PersistentCommand):
             attempt=command.attempts,
         )
         Context.get().update_trace_id('-')
-        SentryReporter.set_tags(command_uuid='-')
+        SentryReporter.reset_command()
 
     def process_timeout(self, e: BaseException):
         LOG.error('Failed with timeout', exc_info=e)
@@ -420,7 +420,7 @@ def seed(self, tenant_uuid: str, recipe_name: str, attempt: int = 0):
         self._execute(tenant_uuid=tenant_uuid)
 
     def _execute(self, tenant_uuid: str):
-        SentryReporter.set_tags(tenant_uuid=tenant_uuid)
+        SentryReporter.set_tags(tenantUuid=tenant_uuid)
         # Run SQL scripts
         app_ctx = Context.get().app
         phase = 'DB'
@@ -487,4 +487,3 @@ def _execute(self, tenant_uuid: str):
                 self.dbs[target].conn_query.connection.commit()
         finally:
             LOG.info('Data seeding done')
-            SentryReporter.set_tags(tenant_uuid='-')

packages/dsw-document-worker/CHANGELOG.md

@@ -17,6 +17,7 @@ to [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
 - `docx-landscape.lua` Pandoc filter (a `landscape` div, or `\landscape` / `\portrait` paragraphs, switch DOCX page orientation)
 
 ### Changed
+- Sentry events carry the same tags as the backend's: `component` is the persistent command component (`document_worker`), and a command sets `function`, `tenantUuid`, `traceUuid` (the command UUID, the `traceId` of the logs), the user who created it, and its trace; the job tags are camelCase (`documentUuid`, `templateUuid`, `formatUuid`) and `command_uuid` / `tenant_uuid` are gone
 
 - Persistent command component (and its notification channel) renamed `doc_worker` → `document_worker`, to match the backend; commands still stored as `doc_worker` are no longer picked up
 - Update to DT metamodel 19.0 (WEP 24): only document contexts of major version 19 are accepted; any other (e.g. of a document queued before the upgrade) fails the document with "Unsupported document context metamodel version X, expected 19.x. Regenerate the document." (`missing` when there is none) and is not reported to Sentry

packages/dsw-document-worker/dsw/document_worker/worker.py

@@ -116,8 +116,8 @@ def get_document(self):
         SentryReporter.set_tags(phase='fetch')
         SentryReporter.set_tags(
             template='?',
-            template_uuid='?',
-            format_uuid='?',
+            templateUuid='?',
+            formatUuid='?',
         )
         if self.tenant_uuid != consts.NULL_UUID:
             LOG.info('Limiting to tenant with UUID: %s', self.tenant_uuid)
@@ -156,8 +156,8 @@ def prepare_template(self):
                  template_uuid, format_uuid)
         # update Sentry info
         SentryReporter.set_tags(
-            template_uuid=template_uuid,
-            format_uuid=format_uuid,
+            templateUuid=template_uuid,
+            formatUuid=format_uuid,
         )
         # prepare template
         template = TemplateRegistry.get().prepare_template(
@@ -402,7 +402,7 @@ def _init_sentry(self):
         SentryReporter.initialize(
             config=self.config.sentry,
             release=BUILD_INFO.version,
-            prog_name=consts.PROG_NAME,
+            component=consts.CMD_COMPONENT,
             event_level=None,
         )
 
@@ -479,38 +479,30 @@ def work(self, command: PersistentCommand):
     @staticmethod
     def _work_pot_file(command: PersistentCommand):
         Context.get().update_trace_id(command.uuid)
-        SentryReporter.set_tags(
-            command_uuid=command.uuid,
-            tenant_uuid=command.tenant_uuid,
-            phase='pot',
-        )
+        SentryReporter.set_command(command)
+        SentryReporter.set_tags(phase='pot')
         LOG.info('Running POT file job #%s', command.uuid)
         PotFileJob(command=command).run()
-        SentryReporter.set_tags(
-            command_uuid='-',
-            tenant_uuid='-',
-            phase='done',
-        )
+        SentryReporter.reset_command()
+        SentryReporter.set_tags(phase='done')
         Context.get().reset_ids()
 
     def _work_document(self, command: PersistentCommand):
         document_uuid = command.body['document']['uuid']
         Context.get().update_trace_id(command.uuid)
         Context.get().update_document_id(document_uuid)
+        SentryReporter.set_command(command)
         SentryReporter.set_tags(
-            command_uuid=command.uuid,
-            tenant_uuid=command.tenant_uuid,
-            document_uuid=document_uuid,
+            documentUuid=document_uuid,
             phase='init',
         )
         LOG.info('Running job #%s', command.uuid)
         self.current_job = Job(command=command, document_uuid=document_uuid)
         self.current_job.run()
         self.current_job = None
+        SentryReporter.reset_command()
         SentryReporter.set_tags(
-            command_uuid='-',
-            tenant_uuid='-',
-            document_uuid='-',
+            documentUuid='-',
             phase='done',
         )
         Context.get().reset_ids()

packages/dsw-mailer/CHANGELOG.md

@@ -14,6 +14,7 @@ to [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
 - The default mail POT (Wizard 5.0): `dsw-mailer make-pot` builds one `mail.pot` of all the templates into `templates/locale/mail.pot` (the image has it), and the mailer keeps it current in S3 at `mail-templates/mail.pot`, with its version as object metadata, at startup and with `dsw-mailer sync-default-pot [--force]` or the Lambda event `{"syncDefaultPot": {...}}`; it never downgrades it unless forced and never blocks mail; `mail.syncDefaultPot` (`MAIL_SYNC_DEFAULT_POT`) turns it off
 
 ### Changed
+- Sentry events carry the same tags as the backend's: `component` is the persistent command component (`mailer`), and a command sets `function`, `tenantUuid`, `traceUuid` (the command UUID, the `traceId` of the logs), the user who created it, and its trace; `command_uuid` / `tenant_uuid` are gone
 
 - Mails are translated with the tenant's mail template locales (Wizard 5.0): one mail is sent per recipient, in the mail template locale whose code matches the recipient's UI locale (exact code, then the base language), otherwise the tenant's default one, otherwise `en`, otherwise built-in English; disabled locales are skipped and a recipient who is not a user gets the default
 - A command with `sendAsOne` sends a single mail to all its recipients in English (the analytics mails); a command without the field, queued before the upgrade, is sent per recipient