feat: S13.21 re-enable @windows_wip store_capacity scenarios

.devcontainer/docker-compose.yml

@@ -43,8 +43,14 @@ services:
           [ "$$SECONDS" -ge "$$deadline" ] && { echo "Timed out waiting for syslog-ng control socket"; kill "$$pid" 2>/dev/null; exit 1; }
           sleep 0.1
         done
-        chmod 0777 /var/lib/syslog-ng/syslog-ng.ctl
-        wait
+        # Watchdog: syslog-ng RELOAD recreates the ctl socket with default 755 perms;
+        # the BDD behave container connects as a non-root user, so keep the socket
+        # world-writable across reloads.
+        ( while kill -0 "$$pid" 2>/dev/null; do
+            chmod 0777 /var/lib/syslog-ng/syslog-ng.ctl 2>/dev/null
+            sleep 0.1
+          done ) &
+        wait $$pid
 
   behave:
     image: ghcr.io/davidcozens/behave:sha-3faff14

Bdd/features/environment.py

@@ -133,10 +133,11 @@ def after_scenario(context, scenario):
                 process.wait()
         del context.interactive_process
 
-    # Restore syslog-ng config if it was changed during the scenario
+    # Restore syslog-ng config if it was changed during the scenario.
+    # copyfile (data only, no chmod) — see syslog_ng_swap_config rationale.
     if hasattr(context, "syslog_ng_config_changed") and context.syslog_ng_config_changed:
-        shutil.copy(SYSLOG_NG_FULL_CONF, SYSLOG_NG_CONF)
         try:
+            shutil.copyfile(SYSLOG_NG_FULL_CONF, SYSLOG_NG_CONF)
             with socket.socket(socket.AF_UNIX, socket.SOCK_STREAM) as sock:
                 sock.connect(SYSLOG_NG_CTL)
                 sock.sendall(b"RELOAD\n")

Bdd/features/steps/syslog_steps.py

@@ -355,13 +355,13 @@ def wait_for_messages(context, expected_messages):
     received_log = context.received_log
     oracle_format = context.oracle_format
     expected_total = context.lines_before + expected_messages
-    deadline = time.monotonic() + 5
+    deadline = time.monotonic() + 10
     while oracle_record_count(received_log, oracle_format) < expected_total:
         if time.monotonic() > deadline:
             actual = oracle_record_count(received_log, oracle_format) - context.lines_before
             raise AssertionError(
                 f"oracle received {actual} of {expected_messages} "
-                f"messages within 5 seconds"
+                f"messages within 10 seconds"
             )
         time.sleep(0.1)
 
@@ -468,8 +468,15 @@ def syslog_ng_reload():
 
 
 def syslog_ng_swap_config(config_path):
-    """Replace the active syslog-ng config and reload."""
-    shutil.copy(config_path, SYSLOG_NG_CONF)
+    """Replace the active syslog-ng config and reload.
+
+    Uses copyfile (data only, no chmod) rather than copy. The destination
+    is bind-mounted from the host on developer machines; chmod fails with
+    'Operation not permitted' when the host owner is not the container's
+    `developer` user. The file already exists with the correct mode
+    so copying just the contents is sufficient.
+    """
+    shutil.copyfile(config_path, SYSLOG_NG_CONF)
     syslog_ng_reload()
 
 
@@ -713,12 +720,17 @@ def step_oracle_stops_tcp(context):
         probe.settimeout(1)
         probe.connect(("syslog-ng", 5514))
 
+        # Set the restore flag *before* the swap so any failure inside
+        # syslog_ng_swap_config (or the subsequent wait/probe-teardown)
+        # still triggers after_scenario's config restore — otherwise a
+        # half-applied swap would leave the on-disk config in UDP-only
+        # mode and contaminate the next scenario.
+        context.syslog_ng_config_changed = True
         syslog_ng_swap_config(SYSLOG_NG_UDP_ONLY_CONF)
         wait_for_tcp_port_closed()
         wait_for_connection_teardown(probe)
         # Allow time for the sender's existing connection to receive RST
         time.sleep(0.5)
-        context.syslog_ng_config_changed = True
     else:
         # Windows OTel: kill the otelcol-contrib process. The TCP listener
         # disappears as the process exits — the resume step will start a

Bdd/features/store_capacity.feature

@@ -4,16 +4,6 @@ Feature: Store capacity limit and discard policy
   When the store is full, the discard policy determines whether
   the oldest or newest messages are dropped.
 
-  Note: scenarios are tagged @windows_wip pending S12.14 (architecture
-  refactor that decouples buffer drain from sender state). The tight-loop
-  `client sends N messages` step fills the in-memory buffer faster than
-  the service thread can drain it on Windows, so records never reach the
-  store and the discard policy can't engage. Linux compose runner happens
-  to schedule producer and service so the buffer stays small enough that
-  the assertions pass; that is timing-dependent and will be re-tuned in
-  the PR #275 finale once S12.14 is on main.
-
-  @windows_wip
   Scenario: Discard-oldest drops oldest messages when store overflows
     Given the syslog oracle is running
     And the block store is enabled with max-blocks 2 and max-block-size 520 and discard-policy oldest
@@ -29,7 +19,6 @@ Feature: Store capacity limit and discard policy
     And the syslog oracle did not receive sequenceId 2
     And the outage messages have contiguous sequenceIds
 
-  @windows_wip
   Scenario: Discard-newest preserves oldest messages when store overflows
     Given the syslog oracle is running
     And the block store is enabled with max-blocks 2 and max-block-size 520 and discard-policy newest
@@ -45,7 +34,6 @@ Feature: Store capacity limit and discard policy
     And the syslog oracle did not receive sequenceId 11
     And the outage messages have contiguous sequenceIds
 
-  @windows_wip
   Scenario: Halt stops the application when store overflows
     Given the syslog oracle is running
     And the block store is enabled with max-blocks 2 and max-block-size 520 and discard-policy halt
@@ -57,7 +45,6 @@ Feature: Store capacity limit and discard policy
     And the client sends 8 messages
     And the client attempts to send it exits with code 2
 
-  @windows_wip
   Scenario: Halt prevents further service after store overflows
     Given the syslog oracle is running
     And the block store is enabled with max-blocks 2 and max-block-size 520 and discard-policy halt