OCI catalog verification, console start marks and log cleanup

oci/remote/oci_console.py

@@ -25,6 +25,17 @@ import sys
 
 NO_LOGIN = ('nologin', 'false')
 LOG_DIR = Path('/var/log/proxmenux/oci')
+# Each start is marked in the console log by a pre-start hook. The hook runs
+# the script only when it exists and always succeeds: a hook that fails would
+# stop the container from starting.
+START_MARK_SCRIPT = Path(__file__).resolve().with_name('oci_console_mark.sh')
+START_MARK = '=== ProxMenux: container started '
+
+
+def start_mark_hook(vmid: int) -> str:
+    script = START_MARK_SCRIPT
+    # `test`, not `[`: a bracket in the configuration reads as a snapshot section.
+    return f"lxc.hook.pre-start: /bin/sh -c 'test -x {script} && {script} {int(vmid)}; exit 0'"
 LOGROTATE = Path('/etc/logrotate.d/proxmenux-oci')
 # copytruncate, because liblxc keeps the file open for as long as the
 # container runs; moving it away would leave the application writing into the
@@ -165,8 +176,9 @@ def enable_log(vmid: int) -> Path:
     text = conf.read_text()
     current, _, snapshots = text.partition('\n[')
     wanted = f'lxc.console.logfile: {path}'
-    kept = [line for line in current.splitlines() if not line.startswith('lxc.console.logfile:')]
-    kept.append(wanted)
+    kept = [line for line in current.splitlines()
+            if not line.startswith('lxc.console.logfile:') and START_MARK_SCRIPT.name not in line]
+    kept += [wanted, start_mark_hook(vmid)]
     rebuilt = '\n'.join(kept) + '\n'
     if snapshots:
         rebuilt += '\n[' + snapshots
@@ -176,6 +188,14 @@ def enable_log(vmid: int) -> Path:
     return path
 
 
+def remove_log(vmid: int) -> None:
+    """Delete the console log of a removed container and its rotated copies."""
+    base = log_path(vmid)
+    for path in [base, *LOG_DIR.glob(f'{base.name}.*')]:
+        if path.is_file() and not path.is_symlink():
+            path.unlink()
+
+
 def configure(vmid: int) -> dict:
     """Console log and Proxmox terminal of a container being created."""
     path = enable_log(vmid)

oci/remote/oci_console_mark.sh

@@ -0,0 +1,9 @@
+#!/bin/sh
+# Marks each start of an OCI container in its console log, so the log can be
+# read from the last start. Run by the container's lxc.hook.pre-start.
+case "$1" in
+  ''|*[!0-9]*) exit 0 ;;
+esac
+printf '\n=== ProxMenux: container started %s ===\n' "$(date '+%Y-%m-%d %H:%M:%S')" \
+  >> "/var/log/proxmenux/oci/$1.console.log" 2>/dev/null
+exit 0

oci/remote/oci_instance_transaction.py

@@ -44,7 +44,9 @@ BASIC = {'arch', 'cmode', 'console', 'tty', 'cores', 'cpulimit', 'cpuunits', 'de
          'lxc.signal.halt', 'lxc.environment.runtime',
          # The container's console log, set by the installer on every
          # creation; the rebuilt container gets it again the same way.
-         'lxc.console.logfile'}
+         'lxc.console.logfile',
+         # The hook that marks each start in that log, set the same way.
+         'lxc.hook.pre-start'}
 # Their output is data (and may hold saved secrets); it is never logged.
 DATA_COMMANDS = {('pct', 'config'), ('pvesh', 'get')}
 LOG_DIR = Path(os.environ.get('OCI_LOG_DIR', '/var/log/proxmenux/oci'))

oci/remote/oci_remove.py

@@ -18,6 +18,7 @@ import sys
 import oci_image_cache as image_cache
 import oci_instances as instances
 from oci_installation_state import parse_config
+import oci_console
 from oci_ui import translate, msg_info, msg_ok, msg_warn, msg_error
 
 # The private networks ProxMenux creates for multi-container applications.
@@ -118,11 +119,13 @@ def remove(root, vmid):
         config = guest_config(member)
         if config is None:
             msg_warn(f"{translate('The container no longer exists:')} CT {member}")
+            oci_console.remove_log(member)
         elif instances.identity(config) != record['installation_id']:
             msg_warn(f"{translate('The VMID belongs to another container now and is not touched:')} CT {member}")
         else:
             subprocess.run(['pct', 'stop', str(member), '--skiplock', '1'], check=False, capture_output=True)
             run('pct', 'destroy', str(member), '--purge', '1', '--destroy-unreferenced-disks', '1')
+            oci_console.remove_log(member)
             msg_ok(f"{translate('Container removed:')} CT {member}")
     if bridge and not bridge_in_use(bridge, set(members)):
         release_bridge(bridge)