feat(notify): Stop using the slack-notifier base (#33928)

.gitlab/internal_kubernetes_deploy/internal_kubernetes_deploy.yml

@@ -2,9 +2,6 @@
 # internal_kubernetes_deploy stage
 # Contains jobs to trigger a pipeline in our k8s-datadog-agent-ops repo
 
-include:
-  - https://gitlab-templates.ddbuild.io/slack-notifier/v3-sdm/template.yml
-
 internal_kubernetes_deploy_experimental:
   stage: internal_kubernetes_deploy
   rules:
@@ -68,7 +65,7 @@ internal_kubernetes_deploy_experimental:
 
 notify-slack:
   stage: internal_kubernetes_deploy
-  extends: .slack-notifier-base
+  image: registry.ddbuild.io/ci/datadog-agent-buildimages/deb_arm64$DATADOG_AGENT_BUILDIMAGES_SUFFIX:$DATADOG_AGENT_BUILDIMAGES
   rules:
     - if: $FORCE_K8S_DEPLOYMENT == "true"
       when: always
@@ -79,9 +76,10 @@ notify-slack:
     - if: $APPS !~ "/^datadog-agent/"
       when: never
     - !reference [.on_deploy]
-  tags: ["arch:amd64"]
+  tags: ["arch:arm64"]
   needs: ["internal_kubernetes_deploy_experimental"]
   script:
     - export SDM_JWT=$(vault read -field=token identity/oidc/token/sdm)
-    - python3 -m pip install -r tasks/requirements.txt
+    - python3 -m pip install -r tasks/requirements.txt -r tasks/libs/requirements-notifications.txt
+    - SLACK_API_TOKEN=$($CI_PROJECT_DIR/tools/ci/fetch_secret.sh $SLACK_AGENT token) || exit $?; export SLACK_API_TOKEN
     - inv pipeline.changelog ${CI_COMMIT_SHORT_SHA} || exit $?

.gitlab/notify/notify.yml

@@ -2,32 +2,35 @@
 # notify stage
 # Contains jobs which send notifications depending on pipeline status.
 
-include:
-  - https://gitlab-templates.ddbuild.io/slack-notifier/v3-sdm/template.yml
+.notify_setup:
+  - SLACK_API_TOKEN=$($CI_PROJECT_DIR/tools/ci/fetch_secret.sh $SLACK_AGENT token) || exit $?; export SLACK_API_TOKEN
+  - GITLAB_TOKEN=$($CI_PROJECT_DIR/tools/ci/fetch_secret.sh $GITLAB_TOKEN read_api) || exit $?; export GITLAB_TOKEN
+  - DD_API_KEY=$($CI_PROJECT_DIR/tools/ci/fetch_secret.sh $AGENT_API_KEY_ORG2 token) || exit $?; export DD_API_KEY
+  - python3 -m pip install -r requirements.txt -r tasks/libs/requirements-notifications.txt
 
 notify-on-tagged-success:
-  extends: .slack-notifier-base
   stage: notify
+  image: registry.ddbuild.io/ci/datadog-agent-buildimages/deb_arm64$DATADOG_AGENT_BUILDIMAGES_SUFFIX:$DATADOG_AGENT_BUILDIMAGES
   rules: !reference [.on_deploy_stable_or_beta_repo_branch]
   dependencies: []
-  tags: ["arch:amd64"]
+  tags: ["arch:arm64"]
   script: |
     MESSAGE_TEXT=":host-green: Tagged build <$CI_PIPELINE_URL|$CI_PIPELINE_ID> succeeded.
     *$CI_COMMIT_REF_NAME* is available in the staging repositories."
-    postmessage "#agent-release-sync" "$MESSAGE_TEXT"
+    python3 -m pip install -r requirements.txt -r tasks/libs/requirements-notifications.txt
+    SLACK_API_TOKEN=$($CI_PROJECT_DIR/tools/ci/fetch_secret.sh $SLACK_AGENT token) || exit $?; export SLACK_API_TOKEN 
+    invoke notify.post-message -c "#agent-release-sync" -m "$MESSAGE_TEXT"
 
 notify:
-  extends: .slack-notifier-base
   stage: notify
+  image: registry.ddbuild.io/ci/datadog-agent-buildimages/deb_arm64$DATADOG_AGENT_BUILDIMAGES_SUFFIX:$DATADOG_AGENT_BUILDIMAGES
   rules: !reference [.on_main_or_release_branch_or_deploy_always]
   dependencies: []
-  tags: ["arch:amd64"]
+  tags: ["arch:arm64"]
   resource_group: notification
   timeout: 15 minutes # Added to prevent a stuck job blocking the resource_group defined above
   script:
-    - GITLAB_TOKEN=$($CI_PROJECT_DIR/tools/ci/fetch_secret.sh $GITLAB_TOKEN read_api) || exit $?; export GITLAB_TOKEN
-    - DD_API_KEY=$($CI_PROJECT_DIR/tools/ci/fetch_secret.sh $AGENT_API_KEY_ORG2 token) || exit $?; export DD_API_KEY
-    - python3 -m pip install -r requirements.txt -r tasks/libs/requirements-notifications.txt
+    - !reference [.notify_setup]
     - invoke -e notify.send-message -p $CI_PIPELINE_ID
     - invoke -e notify.check-consistent-failures -p $CI_PIPELINE_ID
 
@@ -38,8 +41,7 @@ send_pipeline_stats:
   when: always
   dependencies: []
   script:
-    - GITLAB_TOKEN=$($CI_PROJECT_DIR/tools/ci/fetch_secret.sh $GITLAB_TOKEN read_api) || exit $?; export GITLAB_TOKEN
-    - DD_API_KEY=$($CI_PROJECT_DIR/tools/ci/fetch_secret.sh $AGENT_API_KEY_ORG2 token) || exit $?; export DD_API_KEY
+    - !reference [.notify_setup]
     - invoke -e notify.send-stats
 
 notify_github:
@@ -105,11 +107,6 @@ notify_gitlab_ci_changes:
   resource_group: notification
   timeout: 15 minutes # Added to prevent a stuck job blocking the resource_group defined above
 
-.failure_summary_setup:
-  - SLACK_API_TOKEN=$($CI_PROJECT_DIR/tools/ci/fetch_secret.sh $SLACK_AGENT token) || exit $?; export SLACK_API_TOKEN
-  - GITLAB_TOKEN=$($CI_PROJECT_DIR/tools/ci/fetch_secret.sh $GITLAB_TOKEN read_api) || exit $?; export GITLAB_TOKEN
-  - DD_API_KEY=$($CI_PROJECT_DIR/tools/ci/fetch_secret.sh $AGENT_API_KEY_ORG2 token) || exit $?; export DD_API_KEY
-  - python3 -m pip install -r requirements.txt -r tasks/libs/requirements-notifications.txt
 
 # Upload failure summary data to S3 at the end of each main pipeline
 notify_failure_summary_on_pipeline:
@@ -119,7 +116,7 @@ notify_failure_summary_on_pipeline:
       when: never
     - !reference [.on_main_always]
   script:
-    - !reference [.failure_summary_setup]
+    - !reference [.notify_setup]
     - inv -e notify.failure-summary-upload-pipeline-data
 
 # Send failure summary notifications daily and weekly
@@ -130,7 +127,7 @@ notify_failure_summary_daily:
       when: never
     - !reference [.on_deploy_nightly_repo_branch_always]
   script:
-    - !reference [.failure_summary_setup]
+    - !reference [.notify_setup]
     - weekday="$(date --utc '+%A')"
     - |
       if [ "$weekday" = "Sunday" ] || [ "$weekday" = "Monday" ]; then

.gitlab/trigger_release/trigger_release.yml

@@ -77,9 +77,6 @@ trigger_manual_prod_release_on_failure:
       when: never
     - !reference [.on_deploy_stable_on_failure]
 
-include:
-  - https://gitlab-templates.ddbuild.io/slack-notifier/v3-sdm/template.yml
-
 .setup_github_app_agent_platform_auto_pr:
   # GitHub App rate-limits are per-app. Since we are rarely calling the job, we are only using the instance 2
   - |
@@ -90,9 +87,9 @@ include:
 
 generate_windows_gitlab_runner_bump_pr:
   stage: trigger_release
-  extends: .slack-notifier-base
+  image: registry.ddbuild.io/ci/datadog-agent-buildimages/deb_arm64$DATADOG_AGENT_BUILDIMAGES_SUFFIX:$DATADOG_AGENT_BUILDIMAGES
   needs: ["trigger_auto_staging_release"]
-  tags: ["arch:amd64"]
+  tags: ["arch:arm64"]
   rules:
     - if: $DDR_WORKFLOW_ID != null
       when: never
@@ -105,14 +102,15 @@ generate_windows_gitlab_runner_bump_pr:
     - !reference [.setup_github_app_agent_platform_auto_pr]
     - python3 -m pip install -r requirements.txt -r tasks/libs/requirements-notifications.txt
     - $S3_CP_CMD $S3_ARTIFACTS_URI/agent-version.cache .
+    - SLACK_API_TOKEN=$($CI_PROJECT_DIR/tools/ci/fetch_secret.sh $SLACK_AGENT token) || exit $?; export SLACK_API_TOKEN
     - inv -e github.update-windows-runner-version
 
 # Manual job to generate the gitlab bump pr on buildenv if trigger_auto_staging_release fails
 generate_windows_gitlab_runner_bump_pr_manual:
   stage: trigger_release
-  extends: .slack-notifier-base
+  image: registry.ddbuild.io/ci/datadog-agent-buildimages/deb_arm64$DATADOG_AGENT_BUILDIMAGES_SUFFIX:$DATADOG_AGENT_BUILDIMAGES
   needs: ["trigger_auto_staging_release"]
-  tags: ["arch:amd64"]
+  tags: ["arch:arm64"]
   rules:
     - if: $DDR_WORKFLOW_ID != null
       when: never
@@ -125,4 +123,5 @@ generate_windows_gitlab_runner_bump_pr_manual:
     - !reference [.setup_github_app_agent_platform_auto_pr]
     - python3 -m pip install -r requirements.txt -r tasks/libs/requirements-notifications.txt
     - $S3_CP_CMD $S3_ARTIFACTS_URI/agent-version.cache .
+    - SLACK_API_TOKEN=$($CI_PROJECT_DIR/tools/ci/fetch_secret.sh $SLACK_AGENT token) || exit $?; export SLACK_API_TOKEN
     - inv -e github.update-windows-runner-version

tasks/github_tasks.py

@@ -25,7 +25,6 @@
 from tasks.libs.common.datadog_api import create_gauge, send_event, send_metrics
 from tasks.libs.common.git import get_default_branch
 from tasks.libs.common.utils import get_git_pretty_ref
-from tasks.libs.notify.pipeline_status import send_slack_message
 from tasks.libs.owners.linter import codeowner_has_orphans, directory_has_packages_without_owner
 from tasks.libs.owners.parsing import read_owners
 from tasks.libs.pipeline.notifications import GITHUB_SLACK_MAP
@@ -143,7 +142,10 @@ def _update_windows_runner_version(new_version=None, buildenv_ref="master"):
 
     message = f":robobits: A new windows-runner bump PR to {new_version} has been generated. Please take a look :frog-review:\n:pr: {PR_URL} :ty:"
 
-    send_slack_message("ci-infra-support", message)
+    from slack_sdk import WebClient
+
+    client = WebClient(token=os.environ["SLACK_API_TOKEN"])
+    client.chat_postMessage(channel="ci-infra-support", text=message)
     return workflow_conclusion
 
 

tasks/libs/notify/alerts.py

@@ -19,7 +19,6 @@
 from tasks.libs.pipeline.data import get_failed_jobs
 from tasks.libs.pipeline.notifications import (
     get_pr_from_commit,
-    send_slack_message,
 )
 from tasks.owners import channel_owners, make_partition
 
@@ -232,7 +231,10 @@ def send_alert(channel, consecutive: ConsecutiveJobAlert, cumulative: Cumulative
         message = message.strip()
 
         if message:
-            send_slack_message(channel, message)
+            from slack_sdk import WebClient
+
+            client = WebClient(token=os.environ["SLACK_API_TOKEN"])
+            client.chat_postMessage(channel=channel, text=message)
 
             # Create metrics for consecutive and cumulative alerts
             return [

tasks/libs/notify/pipeline_status.py

@@ -1,14 +1,15 @@
+import os
 import re
+import sys
 
 from tasks.libs.ciproviders.gitlab_api import get_commit, get_pipeline
 from tasks.libs.common.git import get_default_branch
+from tasks.libs.common.utils import Color, color_message
 from tasks.libs.notify.utils import DEPLOY_PIPELINES_CHANNEL, PIPELINES_CHANNEL, PROJECT_NAME, get_pipeline_type
 from tasks.libs.pipeline.data import get_failed_jobs
 from tasks.libs.pipeline.notifications import (
     base_message,
-    email_to_slackid,
     get_failed_tests,
-    send_slack_message,
 )
 from tasks.libs.types.types import SlackMessage
 
@@ -58,16 +59,26 @@ def send_message(ctx, pipeline_id, dry_run):
         for test in get_failed_tests(PROJECT_NAME, job):
             message.add_test_failure(test, job)
 
-    # Send messages
+    # Send message
+    from slack_sdk import WebClient
+    from slack_sdk.errors import SlackApiError
+
+    client = WebClient(token=os.environ["SLACK_API_TOKEN"])
     if dry_run:
         print(f"Would send to {slack_channel}:\n{str(message)}")
     else:
-        send_slack_message(slack_channel, str(message))
+        client.chat_postMessage(channel=slack_channel, text=str(message))
 
     if should_send_message_to_author(pipeline.ref, get_default_branch()):
         author_email = commit.author_email
         if dry_run:
             print(f"Would send to {author_email}:\n{str(message)}")
         else:
-            recipient = email_to_slackid(ctx, author_email)
-            send_slack_message(recipient, str(message))
+            try:
+                recipient = client.users_lookupByEmail(email=author_email)
+                client.chat_postMessage(channel=recipient.data['user']['id'], text=str(message))
+            except SlackApiError as e:
+                print(
+                    f"[{color_message('ERROR', Color.RED)}] Failed to send message to {author_email}: {e.response['error']}",
+                    file=sys.stderr,
+                )

tasks/libs/pipeline/notifications.py

@@ -4,14 +4,12 @@
 import os
 import pathlib
 import re
-import subprocess
 from collections import defaultdict
 from datetime import datetime, timezone
 
 import gitlab
 import yaml
 from gitlab.v4.objects import ProjectCommit, ProjectJob, ProjectPipeline
-from invoke.context import Context
 
 from tasks.libs.ciproviders.gitlab_api import get_gitlab_repo
 from tasks.libs.owners.parsing import read_owners
@@ -157,6 +155,9 @@ def base_message(project_name: str, pipeline: ProjectPipeline, commit: ProjectCo
     commit_url_github = f"{GITHUB_BASE_URL}/{project_name}/commit/{commit.id}"
     commit_short_sha = commit.id[-8:]
     author = commit.author_name
+    finish = datetime.fromisoformat(pipeline.finished_at) if pipeline.finished_at else datetime.now(timezone.utc)
+    delta = finish - datetime.fromisoformat(pipeline.started_at)
+    duration = f"[:hourglass: {int(delta.total_seconds() / 60)} min]"
 
     # Try to find a PR id (e.g #12345) in the commit title and add a link to it in the message if found.
     pr_info = get_pr_from_commit(commit_title, project_name)
@@ -165,22 +166,10 @@ def base_message(project_name: str, pipeline: ProjectPipeline, commit: ProjectCo
         parsed_pr_id, pr_url_github = pr_info
         enhanced_commit_title = enhanced_commit_title.replace(f"#{parsed_pr_id}", f"<{pr_url_github}|#{parsed_pr_id}>")
 
-    return f"""{header} pipeline <{pipeline_url}|{pipeline_id}> for {commit_ref_name} {state}.
+    return f"""{header} pipeline <{pipeline_url}|{pipeline_id}> for {commit_ref_name} {state} {duration}.
 {enhanced_commit_title} (<{commit_url_gitlab}|{commit_short_sha}>)(:github: <{commit_url_github}|link>) by {author}"""
 
 
-def send_slack_message(recipient, message):
-    subprocess.run(["postmessage", recipient, message], check=True)
-
-
-def email_to_slackid(ctx: Context, email: str) -> str:
-    slackid = ctx.run(f"echo '{email}' | email2slackid", hide=True, warn=True).stdout.strip()
-
-    assert slackid != '', 'Email not found'
-
-    return slackid
-
-
 def warn_new_commits(release_managers, team, branch, next_rc):
     from slack_sdk import WebClient
 

tasks/notify.py

@@ -277,3 +277,14 @@ def close_failing_tests_stale_issues(_, dry_run=False):
                 print(f'Error closing issue {issue["key"]}: {e}', file=sys.stderr)
 
     print(f'Closed {n_closed} issues without failing tests')
+
+
+@task
+def post_message(_: Context, channel: str, message: str):
+    """
+    Post a message to a slack channel
+    """
+    from slack_sdk import WebClient
+
+    client = WebClient(token=os.environ['SLACK_API_TOKEN'])
+    client.chat_postMessage(channel=channel, text=message)

tasks/pipeline.py

@@ -26,7 +26,6 @@
     is_allowed_repo_branch,
 )
 from tasks.libs.owners.parsing import read_owners
-from tasks.libs.pipeline.notifications import send_slack_message
 from tasks.libs.pipeline.tools import (
     FilteredOutException,
     cancel_pipelines_with_confirmation,
@@ -475,6 +474,9 @@ def is_system_probe(owners, files):
 
 @task
 def changelog(ctx, new_commit_sha):
+    from slack_sdk import WebClient
+
+    client = WebClient(token=os.environ["SLACK_API_TOKEN"])
     # Environment variable to deal with both local and CI environments
     if "CI_PROJECT_DIR" in os.environ:
         parent_dir = os.environ["CI_PROJECT_DIR"]
@@ -502,7 +504,7 @@ def changelog(ctx, new_commit_sha):
     if old_commit_sha == new_commit_sha:
         print("No new commits found, exiting")
         slack_message += no_commits_msg
-        send_slack_message("system-probe-ops", slack_message)
+        client.chat_postMessage(channel="system-probe-ops", text=slack_message)
         return
 
     print(f"Generating changelog for commit range {old_commit_sha} to {new_commit_sha}")
@@ -543,7 +545,7 @@ def changelog(ctx, new_commit_sha):
         slack_message += empty_changelog_msg
 
     print(f"Posting message to slack: \n {slack_message}")
-    send_slack_message("system-probe-ops", slack_message)
+    client.chat_postMessage(channel="system-probe-ops", text=slack_message)
     print(f"Writing new commit sha: {new_commit_sha} to SSM")
     res = ctx.run(
         f"aws ssm put-parameter --name ci.datadog-agent.gitlab_changelog_commit_sha --value {new_commit_sha} "