Update imagerebuild.py for sh 2.0

.idea/misc.xml

@@ -3,7 +3,7 @@
   <component name="Black">
     <option name="sdkName" value="Python 3.11" />
   </component>
-  <component name="ProjectRootManager" version="2" project-jdk-name="Python 3.11" project-jdk-type="Python SDK" />
+  <component name="ProjectRootManager" version="2" project-jdk-name="Python 3.12" project-jdk-type="Python SDK" />
   <component name="PythonCompatibilityInspectionAdvertiser">
     <option name="version" value="3" />
   </component>

.idea/podlaunch.iml

@@ -2,7 +2,7 @@
 <module type="PYTHON_MODULE" version="4">
   <component name="NewModuleRootManager">
     <content url="file://$MODULE_DIR$" />
-    <orderEntry type="jdk" jdkName="Python 3.11" jdkType="Python SDK" />
+    <orderEntry type="jdk" jdkName="Python 3.12" jdkType="Python SDK" />
     <orderEntry type="sourceFolder" forTests="false" />
   </component>
 </module>

imagerebuild.py

@@ -8,11 +8,17 @@ from typing import Set, Callable
 
 import click
 import sh
-from sh import podman
-from sh import systemctl
 
 SERVICES_BASE_PATH = "/infra/services/"
 
+progressbar = click.progressbar
+
+
+def hidden_progressbar(*args, **kwargs):
+    bar = click.progressbar(*args, **kwargs)
+    bar.is_hidden = True
+    return bar
+
 
 def resolve_image_units():
     services_path = pathlib.Path(SERVICES_BASE_PATH)
@@ -20,33 +26,29 @@ def resolve_image_units():
 
     logging.info(f"Found {len(services_set)} services: {str(services_set)}")
 
-    systemctl("daemon-reload")
+    sh.systemctl("daemon-reload")
 
     def remove_masked_unit(
         _item_set: Set[str],
         item: str,
         item_to_unit: Callable[[str], str] = lambda i: i,
     ):
-        load_state = systemctl.show(
+        load_state_output = sh.systemctl.show(
             "--property=LoadState", "--value", item_to_unit(item)
         )
-        load_state = load_state.stdout.strip().decode(
-            encoding="utf-8", errors="replace"
-        )
+        load_state = load_state_output.strip()
         logging.debug(f"{item} load state: {repr(load_state)}")
         if load_state == "masked":
             logging.info(f"Removed masked entry: {item}")
             _item_set.remove(item)
 
-    with click.progressbar(list(services_set), label="Checking service units..", show_pos=True) as bar:
+    with progressbar(list(services_set), label="Checking service units..", show_pos=True) as bar:
         for service in bar:
             remove_masked_unit(services_set, service, lambda srv: f"pod@{srv}.service")
 
     def add_wants_to_image_units(_image_units: Set[str], unit: str):
-        wants = systemctl.show("--property=Wants", "--value", unit)
-        wants_list = (
-            wants.stdout.strip().decode(encoding="utf-8", errors="replace").split(" ")
-        )
+        wants_output = sh.systemctl.show("--property=Wants", "--value", unit)
+        wants_list = wants_output.strip().split(" ")
         logging.debug(f"{unit} wants: {repr(wants_list)}")
         for next_unit in wants_list:
             if next_unit.startswith("image@") and next_unit.endswith(".service"):
@@ -55,7 +57,7 @@ def resolve_image_units():
 
     image_units: Set[str] = set()
 
-    with click.progressbar(
+    with progressbar(
         length=len(services_set) * 2, label="Collecting container image services.."
     ) as bar:
         for service in services_set:
@@ -77,7 +79,7 @@ def resolve_image_units():
                 new_image_units
             )  # add new image units to all image units
 
-    with click.progressbar(
+    with progressbar(
         list(image_units), label="Checking container image units..", show_pos=True
     ) as bar:
         for image_unit in bar:
@@ -88,39 +90,47 @@ def resolve_image_units():
 
 
 @click.command()
-@click.option("--verbose", is_flag=True, default=False, help="Enable INFO logging")
-def main(verbose):
-    if verbose:
+@click.option("-v", "--verbose", count=True, help="Enable INFO logging, use twice for DEBUG")
+def main(verbose: int):
+    if verbose > 0:
+        if verbose > 2:
+            logging.root.setLevel(logging.DEBUG)
+        elif verbose == 2:
+            logging.root.setLevel(logging.DEBUG)
+            shlogger = logging.getLogger("sh")
+            shlogger.setLevel(logging.INFO)
+        else:
             logging.root.setLevel(logging.INFO)
 
+        global progressbar
+        progressbar = hidden_progressbar
+
     image_units = resolve_image_units()
     image_tags: Set[str] = set()
 
-    with click.progressbar(image_units, label="Collecting container image tags..") as bar:
+    with progressbar(image_units, label="Collecting container image tags..") as bar:
         for image_unit in bar:
-            environment = systemctl.show(
+            environment_output = sh.systemctl.show(
                 "--property=Environment",
                 "--value",
                 image_unit,
             )
-            environment_list = (
-                environment.stdout.strip()
-                .decode(encoding="utf-8", errors="replace")
-                .split(" ")
-            )
+            environment_list = environment_output.strip().split(" ")
             logging.debug(f"{image_unit} environment: {repr(environment_list)}")
             for envvar in environment_list:
                 search_str = "IMAGE_TAG="
                 if envvar.startswith(search_str):
-                    image_tags.add(envvar[len(search_str) :])
+                    keylen = len(search_str)
+                    image_tags.add(envvar[keylen:])
 
     started_processes = []
-    with click.progressbar(
+    with progressbar(
         length=len(image_tags), label="Untagging container images..", show_pos=True
     ) as bar:
         for image_tag in image_tags:
-            process = podman.untag(
+            process = sh.podman.untag(
                 image_tag,
+                _return_cmd=True,
                 _bg=True,
                 _err_to_out=True,
                 _done=lambda cmd, success, exit_code: bar.update(1),
@@ -131,7 +141,7 @@ def main(verbose):
         for p in started_processes:
             try:
                 p.wait()
-            except sh.ErrorReturnCode as error:
+            except sh.ErrorReturnCode:
                 # ignore missing image tags
                 if "image not known".encode() in p.stdout:
                     pass
@@ -139,18 +149,16 @@ def main(verbose):
                     raise
 
     started_processes = []
-    with click.progressbar(
+    with progressbar(
         length=len(image_units), label="Building images..", show_pos=True
     ) as bar:
         semaphore = multiprocessing.Semaphore(8)
         for image_unit in image_units:
             try:
-                systemctl("reset-failed", image_unit, _bg=False, _err_to_out=True)
+                sh.systemctl("reset-failed", image_unit, _bg=False, _err_to_out=True)
             except sh.ErrorReturnCode as error:
                 if f"Unit {image_unit} not loaded".encode() in error.stdout:
-                    logging.info(
-                        f"Not resetting failed state for {image_unit}, unit not loaded"
-                    )
+                    logging.info(f"Not resetting failed state for {image_unit}, unit not loaded")
                 else:
                     raise
 
@@ -162,7 +170,9 @@ def main(verbose):
                     logging.warning(f"{cmd.cmd}{tuple(cmd.call_args)} completed with exit code {exit_code}")
                 semaphore.release()
 
-            process = systemctl.restart(image_unit, _bg=True, _done=restart_done)
+            process = sh.systemctl.restart(
+                image_unit, _return_cmd=True, _bg=True, _done=restart_done
+            )
             started_processes.append(process)
         # join processes
         [p.wait() for p in started_processes]

main.py

@@ -1,11 +1,12 @@
+import datetime
 import json
 import logging
 import os
 import pathlib
+import random
 import sys
 import threading
 import traceback
-from datetime import datetime, timedelta
 from queue import SimpleQueue
 from signal import signal, SIGHUP, SIGINT, SIGTERM, setitimer, SIGALRM, ITIMER_REAL, SIGUSR1, SIGUSR2, strsignal
 
@@ -13,6 +14,8 @@ import click
 import sh
 
 SERVICES_BASE_PATH = "/infra/services/"
+POD_CHECK_TIME = 120.0
+RAND_OFFSET_MAX_SECONDS = 10.0
 
 shlog = sh.bake(_out=sys.stdout, _err=sys.stderr)
 sdnotify = sh.Command("systemd-notify")
@@ -41,7 +44,7 @@ class PodKeeper:
         self.reloading = threading.Event()
         self.checking = threading.Event()
         self.waiter = threading.Event()
-        self.last_check = datetime.utcnow()
+        self.last_check = datetime.datetime.now(datetime.UTC)
         self.passing_signal = threading.Event()
         self.pass_signal_nums = SimpleQueue()
 
@@ -115,14 +118,20 @@ class PodKeeper:
             traceback.print_exc()
 
     def check_pod(self):
-        new_timestamp = datetime.utcnow()
+        new_timestamp = datetime.datetime.now(datetime.UTC)
         inspect_command = sh.podman.pod.inspect(self.podname, _return_cmd=True)
-        pod_description = json.loads(inspect_command.stdout)
+        multiple_descriptions = json.loads(inspect_command.stdout)
+        if not multiple_descriptions:
+            print(f"No pod descriptions found for {self.podname}", file=sys.stderr, flush=True)
+            self.stopping.set()
+            return
+        assert len(multiple_descriptions) == 1, f"Single pod description expected for {self.podname}"
+        pod_description = multiple_descriptions[0]
         for container in pod_description["Containers"]:
             if container["State"] != "running":
                 print(f"Container {container['Name']} exited", file=sys.stderr, flush=True)
-                logs_since = self.last_check - timedelta(seconds=10)
-                print(f"Log since last check (-10s):\n", file=sys.stderr, flush=True)
+                logs_since = self.last_check - datetime.timedelta(seconds=POD_CHECK_TIME)
+                print(f"Log since last check (-{POD_CHECK_TIME}s):\n", file=sys.stderr, flush=True)
                 shlog.podman.logs('--since', logs_since.isoformat(), container['Name'], _out=sys.stderr)
                 self.stopping.set()
         self.last_check = new_timestamp
@@ -178,7 +187,8 @@ def main(network, log_driver, log_level, replace, remove, verbose, identifier):
     signal(SIGALRM, keeper.check)
     signal(SIGUSR1, keeper.passthrough)
     signal(SIGUSR2, keeper.passthrough)
-    setitimer(ITIMER_REAL, 4.0, 120.0)
+    random_offset = random.random() * RAND_OFFSET_MAX_SECONDS
+    setitimer(ITIMER_REAL, 1.0 + random_offset, POD_CHECK_TIME + random_offset)
 
     keeper.run()