Add automated retries on retryable condition for building images in CI (

apache#24006)

There is a flakiness in pushing cache images to ghcr.io, therefore
we want to add automated retries when the images fail intermittently.

The root cause of the problem is tracked in containerd:
containerd/containerd#5978

.github/workflows/ci.yml

@@ -1689,6 +1689,7 @@ ${{ hashFiles('.pre-commit-config.yaml') }}"
         run: >
           breeze build-image
           --prepare-buildx-cache
+          --max-retries 3
           --platform linux/amd64,linux/arm64
         env:
           PYTHON_MAJOR_MINOR_VERSION: ${{ matrix.python-version }}
@@ -1722,6 +1723,7 @@ ${{ hashFiles('.pre-commit-config.yaml') }}"
           --install-packages-from-context
           --prepare-buildx-cache
           --disable-airflow-repo-cache
+          --max-retries 3
           --platform linux/amd64,linux/arm64
         env:
           PYTHON_MAJOR_MINOR_VERSION: ${{ matrix.python-version }}

dev/breeze/src/airflow_breeze/commands/ci_image_commands.py

@@ -18,6 +18,7 @@
 import os
 import sys
 from pathlib import Path
+from subprocess import CompletedProcess
 from typing import List, Optional, Tuple, Union
 
 import click
@@ -50,6 +51,7 @@
     option_image_name,
     option_image_tag,
     option_install_providers_from_sources,
+    option_max_retries,
     option_parallelism,
     option_platform,
     option_prepare_buildx_cache,
@@ -87,6 +89,7 @@
     instruct_build_image,
     is_repo_rebased,
     run_command,
+    run_result_contains,
 )
 
 CI_IMAGE_TOOLS_COMMANDS = {
@@ -110,6 +113,7 @@
                 "--tag-as-latest",
                 "--docker-cache",
                 "--force-build",
+                "--max-retries",
             ],
         },
         {
@@ -202,6 +206,7 @@
 @option_docker_cache
 @option_image_tag
 @option_prepare_buildx_cache
+@option_max_retries
 @option_push_image
 @option_empty_image
 @option_install_providers_from_sources
@@ -477,15 +482,34 @@ def build_ci_image(verbose: bool, dry_run: bool, ci_image_params: BuildCiParams)
         )
     else:
         get_console().print(f"\n[info]Building CI Image for Python {ci_image_params.python}\n")
-        build_command_result = run_command(
-            cmd, verbose=verbose, dry_run=dry_run, cwd=AIRFLOW_SOURCES_ROOT, text=True, check=False
-        )
-        if build_command_result.returncode == 0:
-            if ci_image_params.prepare_buildx_cache:
+        num_tries = 1 if ci_image_params.max_retries is None else ci_image_params.max_retries
+        build_command_result = CompletedProcess(args=[], returncode=1, stdout="This should never happen.")
+        while num_tries > 0:
+            build_command_result = run_command(
+                cmd,
+                verbose=verbose,
+                dry_run=dry_run,
+                cwd=AIRFLOW_SOURCES_ROOT,
+                check=False,
+                text=True,
+                capture_output=True,
+            )
+            if ci_image_params.prepare_buildx_cache and build_command_result.returncode == 0:
                 build_command_result = build_cache(
                     image_params=ci_image_params, dry_run=dry_run, verbose=verbose
                 )
-
+            if build_command_result.returncode == 0:
+                break
+            num_tries -= 1
+            if run_result_contains(build_command_result, "cannot reuse body, request must be retried"):
+                if num_tries > 0:
+                    get_console().print(
+                        "[info]Retrying failed command on retryable condition. "
+                        f"There are {num_tries} left[/]"
+                    )
+                continue
+            else:
+                break
     if not ci_image_params.prepare_buildx_cache:
         if not dry_run:
             if build_command_result.returncode == 0:
@@ -504,7 +528,9 @@ def build_ci_image(verbose: bool, dry_run: bool, ci_image_params: BuildCiParams)
                     f"Image build: {ci_image_params.python}",
                 )
         else:
-            get_console().print("[info]Not updating build cache because we are in `dry_run` mode.[/]")
+            get_console().print(
+                "[info]Not tagging/marking image as refreshed because we are in `dry_run` mode.[/]"
+            )
     return build_command_result.returncode, f"Image build: {ci_image_params.python}"
 
 

dev/breeze/src/airflow_breeze/commands/production_image_commands.py

@@ -17,6 +17,7 @@
 import contextlib
 import os
 import sys
+from subprocess import CompletedProcess
 from typing import Optional, Tuple
 
 import click
@@ -49,6 +50,7 @@
     option_image_name,
     option_image_tag,
     option_install_providers_from_sources,
+    option_max_retries,
     option_parallelism,
     option_platform,
     option_prepare_buildx_cache,
@@ -77,7 +79,12 @@
 from airflow_breeze.utils.python_versions import get_python_version_list
 from airflow_breeze.utils.registry import login_to_github_docker_registry
 from airflow_breeze.utils.run_tests import verify_an_image
-from airflow_breeze.utils.run_utils import filter_out_none, fix_group_permissions, run_command
+from airflow_breeze.utils.run_utils import (
+    filter_out_none,
+    fix_group_permissions,
+    run_command,
+    run_result_contains,
+)
 
 PRODUCTION_IMAGE_TOOLS_COMMANDS = {
     "name": "Production Image tools",
@@ -99,6 +106,7 @@
                 "--image-tag",
                 "--tag-as-latest",
                 "--docker-cache",
+                "--max-retries",
             ],
         },
         {
@@ -206,6 +214,7 @@
 @option_docker_cache
 @option_image_tag
 @option_prepare_buildx_cache
+@option_max_retries
 @option_push_image
 @option_empty_image
 @option_airflow_constraints_mode_prod
@@ -517,16 +526,36 @@ def build_production_image(
             image_params=prod_image_params,
             verbose=verbose,
         )
-        build_command_result = run_command(
-            cmd, verbose=verbose, dry_run=dry_run, cwd=AIRFLOW_SOURCES_ROOT, check=False, text=True
-        )
-        if build_command_result.returncode == 0:
-            if prod_image_params.prepare_buildx_cache:
-                build_command_result = build_cache(
-                    image_params=prod_image_params, dry_run=dry_run, verbose=verbose
-                )
+        num_tries = 1 if prod_image_params.max_retries is None else prod_image_params.max_retries
+        build_command_result = CompletedProcess(args=[], returncode=1, stdout="This should never happen.")
+        while num_tries > 0:
+            build_command_result = run_command(
+                cmd,
+                verbose=verbose,
+                dry_run=dry_run,
+                cwd=AIRFLOW_SOURCES_ROOT,
+                check=False,
+                text=True,
+                capture_output=True,
+            )
+            if build_command_result.returncode == 0:
+                if prod_image_params.prepare_buildx_cache:
+                    build_command_result = build_cache(
+                        image_params=prod_image_params, dry_run=dry_run, verbose=verbose
+                    )
+                else:
+                    if prod_image_params.tag_as_latest:
+                        build_command_result = tag_image_as_latest(prod_image_params, dry_run, verbose)
+            if build_command_result.returncode == 0:
+                break
+            num_tries -= 1
+            if run_result_contains(build_command_result, "cannot reuse body, request must be retried"):
+                if num_tries > 0:
+                    get_console().print(
+                        "[info]Retrying failed command on retryable condition. "
+                        f"There are {num_tries} left[/]"
+                    )
+                continue
             else:
-                if prod_image_params.tag_as_latest:
-                    build_command_result = tag_image_as_latest(prod_image_params, dry_run, verbose)
-
+                break
     return build_command_result.returncode, f"Image build: {prod_image_params.python}"

dev/breeze/src/airflow_breeze/params/_common_build_params.py

@@ -55,6 +55,7 @@ class _CommonBuildParams:
     github_token: str = os.environ.get('GITHUB_TOKEN', "")
     github_username: str = ""
     image_tag: Optional[str] = None
+    max_retries: Optional[int] = None
     install_providers_from_sources: bool = False
     platform: str = f"linux/{os.uname().machine}"
     prepare_buildx_cache: bool = False

dev/breeze/src/airflow_breeze/utils/common_options.py

@@ -289,6 +289,12 @@
     is_flag=True,
     envvar='PREPARE_BUILDX_CACHE',
 )
+option_max_retries = click.option(
+    '--max-retries',
+    help='Maximum number of retries for the operation for "retryable" intermittent problems.',
+    type=click.IntRange(min=2),
+    envvar='MAX_RETRIES',
+)
 option_push_image = click.option(
     '--push-image',
     help='Push image after building it.',

dev/breeze/src/airflow_breeze/utils/run_utils.py

@@ -405,3 +405,11 @@ def get_runnable_ci_image(verbose: bool, dry_run: bool) -> str:
         instruction=f"breeze build-image --python {python_version}",
     )
     return airflow_image
+
+
+def run_result_contains(result: RunCommandResult, message: str) -> bool:
+    if result.stdout and message in result.stdout:
+        return True
+    if result.stderr and message in result.stderr:
+        return True
+    return False