-
Notifications
You must be signed in to change notification settings - Fork 16.4k
Modify breeze commands and flags for standalone shell commands #35862
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Conversation
|
This is a logically-separated PR from #35830 as discussed in #35830 (review) It's mostly revolving about making I still left a few minor "loosely-related" fixes. I could probably extract out a few of those in separate PRs if you think it makes sense ("Few other cleanups and minor fixes were also implemented as part of this change" above). But it would inflate number of PRs and since it is just But if needed I can separate them. |
c77b1af to
e06cae0
Compare
|
BTW. Even with those small fixes adde - it should be FAR easier to review than #35830 |
e06cae0 to
d23cafe
Compare
Breeze `shell` command has been foreseen eventually to also allow
single commands in CI and pre-commit, and we are preparing it to
be able to do so with all the bells and whistles. There are a
few improvements needed to be able to do it smoothly.
There are a few things that were preventing using `breeze shell` as
general-purpose commmand executor:
* we used defailt `docker-compose` project for running compose runs
(derived from `docker-compose` parent folder of the `basy.yml` compose
file. If we want to run standalone commands, we need to run them in
separate projects in order to allow to run them while `breeze` shell
is already running
* when runing command in pre-commit hook, there is no terminal available
because this command is run deep under-the-hood in git internals and
it is run as command detached from the terminal it runs in. For that
automated terminal allocation strategy used by docker compose was
not working
* breeze prints a lot of diagnostics of what's going on when you run
the entrypoint by default - including diagnostics information
printed by docker compose and automated docker builds. This is
polluting the output of automated scripts (especially in pre-commits)
* breeze command checks by defalt if image needs to be upgraded and
prints information about it and allows the user to trigger the
upgrade by answering "yes" for 10 seconds. This add extra delay
and unnecessary information when running the scripts in pre-commits
and generally in unattended mode.
* Breeze CI entrypoint uses additional environment initialization
when entering the container - it waits until DBs are available
and performs few other initializaiton checks (for example
loading scripts from `init.sh`. All that is not needed and should
be skipped when running standalone scripts.
* Sometimes it's not obious what internal commands are run by breeze
shell - and it's difficult to debug/diagnose it. We run CI commands
with `--verbose` flags so those command should not be printed even
if `--verbose` flag is used though as they are polluting the output.
* By default some flags are persistently stored after used (python
version, backend, backend version) and they are "cached" - they
are automatically used next time you start breeze - to keep the
environment you use for the last time. Those shoudl not be
persisted when running standalone command and want to override Python
version or backend only in the command you want to execute.
The PR implements the following changes to improve those aspects:
* Allow specifying `--project-name` when running `breeze`, `breze shell`
`breeze start-airflow` and `breeze down` commands. Default project is
`breeze` now and user can use any project but predefined ones are
also `pre-commit` (to be used in pre-commits) and legacy
`docker-compose` if one would like to use `breeze down` to shut down
some remaining containers from the legacy project.
* Add `--restart` (aliased by `--remove-orphans`) flag that might be
used when starting the command, which will also remove orphan
containers (for example postgres container running while we are
switching to sqlite).
* Add `--tty` option ("auto", "disabled", "enabled") in `breeze shell`
command. This flag allows to determine strategy of terminal allocation
by the underlying docker-compose flags.
* Add `--quiet` flag that silences the output of entrypoint diagnostics
including the output of docker-compose and docker build when they
are running as part of the command.
NOTE: We also need to upgrade min docker-compose and docker version
as we are using features available only in recent versions of these
that allow to suppress some output.
* Add `--skip-image-upgrade-check` in `breeze shell` to avoid checking
if upgrade is needed. However if the image is not present, the latest
version will be upgraded anyway - this allows to run pre-commits even
if breeze has not been used before or `docker system prune -a`
was used. We had `--skip-image-check` before, but we rename it to
explain what check we are talking about.
* Add `--skip-environment-initialization` in `breeze shell` to skip
all initialization happening for interactive breeze entering.
* Add `--verbose-commands` flag in `breeze shell`. For a long time
we had `VERBOSE_COMMANDS` env variable support doing it (and printing
commands as they are executed) but adding it explicitly as flag makes
it easier discoverable and usable.
* Add `SKIP_SAVING_CHOICES` environment variable. When this variable
is set, the choices you specify when runing breeze command are
not stored in cache and only used for the particular command you
execute.
Few other cleanups and minor fixes were also implemented as part of this
change:
* the click decorators and corresponding parameters in
breeze/shell/start-airflow commands have been sorted to easier manage
them (adding/updating the option became a real pain for those as they
have a lot of them)
* version_suffix_for_pypi default for CI image has been set up to dev0
to make automated upgrade after pressing `Y` use the same cache
as `docker ci-image build`
* added PYTHONWARNINGS to be forwardeable from host environment to
container - which allows to control which warnings to display.
* only print "Github Actions" group end when we are running inside
GitHub Actions.
* switch from distutls Versions to packaging Version - used
to check pre-commit version (better Python 3.12 compatibility)
* clarified that executor flag for "shell" command can accept more
values than the one in "start-airflow"
* when config options are not synchronized with "rich-click" help
groups, proper location is displayed in error messsages (it asked
to fix the issue in non-existing files).
d23cafe to
3930c81
Compare
jscheffl
left a comment
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
First I was fearing the size to review - until I realized that 50% was boilerplate code changes and 50% SVG image generated :-D LGTM!
Precisely :). |
When pre-commits are run and breeze shell is opened they were both attempting to use the same port forwarding. However pre-commits do not ever need port forwarding because the forwarded ports are only needed when you run breeze shell interactively. This is a follow-up afte apache#35862. This PR removes port forwarding when "pre-commit" project name is used to run breeze shell command.
…5922) When pre-commits are run and breeze shell is opened they were both attempting to use the same port forwarding. However pre-commits do not ever need port forwarding because the forwarded ports are only needed when you run breeze shell interactively. This is a follow-up afte #35862. This PR removes port forwarding when "pre-commit" project name is used to run breeze shell command.
So far we executed commands in CI image in breeze in two ways: * entering the shell (which runs docker-compose under the hood) * running `docker run` with the CI image This requires rather complex mapping of environment variables between `docker-compose` and `docker`. Since recently (apache#35862) we can use `shell` command to run commands in very similar way as docker run (with docker-compose, without database and extra components - just using the same `breeze shell` mechanisms. This PR converts all the usages of docker run CI_IMAGE we had and converts them to use modified `enter_shell` method that has been moved to "docker_command_utils". This also simplified passing arguments to the "enter_shell" command - no longer need to filter out none parameters and **kwargs - all parameters are passed explicitly. This also allowed to remove some of the code (extracting args, filtering_out_none) that are not used anymore. The entypoint CI has been slightly refactored - to provide a bit better structure and handle `--skip-environment-initialization` better - we can now both set `--use-airflow-version` and `--skip-environment-initialization` which was not possible before.
So far we executed commands in CI image in breeze in two ways: * entering the shell (which runs docker-compose under the hood) * running `docker run` with the CI image This requires rather complex mapping of environment variables between `docker-compose` and `docker`. Since recently (#35862) we can use `shell` command to run commands in very similar way as docker run (with docker-compose, without database and extra components - just using the same `breeze shell` mechanisms. This PR converts all the usages of docker run CI_IMAGE we had and converts them to use modified `enter_shell` method that has been moved to "docker_command_utils". This also simplified passing arguments to the "enter_shell" command - no longer need to filter out none parameters and **kwargs - all parameters are passed explicitly. This also allowed to remove some of the code (extracting args, filtering_out_none) that are not used anymore. The entypoint CI has been slightly refactored - to provide a bit better structure and handle `--skip-environment-initialization` better - we can now both set `--use-airflow-version` and `--skip-environment-initialization` which was not possible before.
…5922) When pre-commits are run and breeze shell is opened they were both attempting to use the same port forwarding. However pre-commits do not ever need port forwarding because the forwarded ports are only needed when you run breeze shell interactively. This is a follow-up afte #35862. This PR removes port forwarding when "pre-commit" project name is used to run breeze shell command. (cherry picked from commit ad04d2a)
So far we executed commands in CI image in breeze in two ways: * entering the shell (which runs docker-compose under the hood) * running `docker run` with the CI image This requires rather complex mapping of environment variables between `docker-compose` and `docker`. Since recently (#35862) we can use `shell` command to run commands in very similar way as docker run (with docker-compose, without database and extra components - just using the same `breeze shell` mechanisms. This PR converts all the usages of docker run CI_IMAGE we had and converts them to use modified `enter_shell` method that has been moved to "docker_command_utils". This also simplified passing arguments to the "enter_shell" command - no longer need to filter out none parameters and **kwargs - all parameters are passed explicitly. This also allowed to remove some of the code (extracting args, filtering_out_none) that are not used anymore. The entypoint CI has been slightly refactored - to provide a bit better structure and handle `--skip-environment-initialization` better - we can now both set `--use-airflow-version` and `--skip-environment-initialization` which was not possible before. (cherry picked from commit 2d15dbf)
Breeze
shellcommand has been foreseen eventually to also allow single commands in CI and pre-commit, and we are preparing it to be able to do so with all the bells and whistles. There are a few improvements needed to be able to do it smoothly.There are a few things that were preventing using
breeze shellas general-purpose commmand executor:we used defailt
docker-composeproject for running compose runs (derived fromdocker-composeparent folder of thebasy.ymlcompose file. If we want to run standalone commands, we need to run them in separate projects in order to allow to run them whilebreezeshell is already runningwhen runing command in pre-commit hook, there is no terminal available because this command is run deep under-the-hood in git internals and it is run as command detached from the terminal it runs in. For that automated terminal allocation strategy used by docker compose was not working
breeze prints a lot of diagnostics of what's going on when you run the entrypoint by default - including diagnostics information printed by docker compose and automated docker builds. This is polluting the output of automated scripts (especially in pre-commits)
breeze command checks by defalt if image needs to be upgraded and prints information about it and allows the user to trigger the upgrade by answering "yes" for 10 seconds. This add extra delay and unnecessary information when running the scripts in pre-commits and generally in unattended mode.
Breeze CI entrypoint uses additional environment initialization when entering the container - it waits until DBs are available and performs few other initializaiton checks (for example loading scripts from
init.sh. All that is not needed and should be skipped when running standalone scripts.Sometimes it's not obious what internal commands are run by breeze shell - and it's difficult to debug/diagnose it. We run CI commands with
--verboseflags so those command should not be printed even if--verboseflag is used though as they are polluting the output.By default some flags are persistently stored after used (python version, backend, backend version) and they are "cached" - they are automatically used next time you start breeze - to keep the environment you use for the last time. Those shoudl not be persisted when running standalone command and want to override Python version or backend only in the command you want to execute.
The PR implements the following changes to improve those aspects:
Allow specifying
--project-namewhen runningbreeze,breze shellbreeze start-airflowandbreeze downcommands. Default project isbreezenow and user can use any project but predefined ones are alsopre-commit(to be used in pre-commits) and legacydocker-composeif one would like to usebreeze downto shut down some remaining containers from the legacy project.Add
--restart(aliased by--remove-orphans) flag that might be used when starting the command, which will also remove orphan containers (for example postgres container running while we are switching to sqlite).Add
--ttyoption ("auto", "disabled", "enabled") inbreeze shellcommand. This flag allows to determine strategy of terminal allocation by the underlying docker-compose flags.Add
--quietflag that silences the output of entrypoint diagnostics including the output of docker-compose and docker build when they are running as part of the command.NOTE: We also need to upgrade min docker-compose and docker version as we are using features available only in recent versions of these that allow to suppress some output.
Add
--skip-image-upgrade-checkinbreeze shellto avoid checking if upgrade is needed. However if the image is not present, the latest version will be upgraded anyway - this allows to run pre-commits even if breeze has not been used before ordocker system prune -awas used. We had--skip-image-checkbefore, but we rename it to explain what check we are talking about.Add
--skip-environment-initializationinbreeze shellto skip all initialization happening for interactive breeze entering.Add
--verbose-commandsflag inbreeze shell. For a long time we hadVERBOSE_COMMANDSenv variable support doing it (and printing commands as they are executed) but adding it explicitly as flag makes it easier discoverable and usable.Add
SKIP_SAVING_CHOICESenvironment variable. When this variable is set, the choices you specify when runing breeze command are not stored in cache and only used for the particular command you execute.Few other cleanups and minor fixes were also implemented as part of this change:
the click decorators and corresponding parameters in breeze/shell/start-airflow commands have been sorted to easier manage them (adding/updating the option became a real pain for those as they have a lot of them)
version_suffix_for_pypi default for CI image has been set up to dev0 to make automated upgrade after pressing
Yuse the same cache asdocker ci-image buildadded PYTHONWARNINGS to be forwardeable from host environment to container - which allows to control which warnings to display.
only print "Github Actions" group end when we are running inside GitHub Actions.
switch from distutls Versions to packaging Version - used to check pre-commit version (better Python 3.12 compatibility)
clarified that executor flag for "shell" command can accept more values than the one in "start-airflow"
when config options are not synchronized with "rich-click" help groups, proper location is displayed in error messsages (it asked to fix the issue in non-existing files).
^ Add meaningful description above
Read the Pull Request Guidelines for more information.
In case of fundamental code changes, an Airflow Improvement Proposal (AIP) is needed.
In case of a new dependency, check compliance with the ASF 3rd Party License Policy.
In case of backwards incompatible changes please leave a note in a newsfragment file, named
{pr_number}.significant.rstor{issue_number}.significant.rst, in newsfragments.