Harden schedule bump against Slack retries

The message listener runs synchronously before the 200 ack, so a slow
delete+repost can blow Slack's 3s window and be retried with no guard.
Ignore retried deliveries (X-Slack-Retry-Num), stamp last_bump_ts before
the delete/repost so a mid-bump retry is debounced, and skip thread
replies that don't move the main timeline.

Refs: #135

docs/sticky-schedule-post.md

@@ -101,15 +101,23 @@ request URL. On a `message.channels` event in the schedule channel
   stored as `last_bump_ts` on the schedule-post record, so the debounce check
   reads no channel history.
 - **Idempotent / safe** — the bump is skipped when:
+  - the delivery is a Slack **retry** (`X-Slack-Retry-Num` set) — under
+    `process_before_response=True` the delete+repost runs before the 200 ack, so
+    a slow run can be retried; we never bump on a retry (the next real message
+    bumps anyway),
+  - the triggering message is a **thread reply** (`thread_ts`) — a threaded reply
+    doesn't push the schedule down the main timeline,
   - there is no stored post,
   - the triggering message is the bot's own (`bot_id`) or a non-user message
     subtype (edits, deletes, joins, …),
   - the debounce window has not elapsed, or
   - the bot isn't in the channel (`not_in_channel`).
-- Otherwise it `chat_delete`s the stored post and `chat_postMessage`s the current
-  `build_week_schedule` blocks, then `save_schedule_post`s the new `ts` (with
-  `last_bump_ts`). If the old message is already gone, it still reposts so the
-  channel always ends with the schedule.
+- Otherwise it **stamps `last_bump_ts` optimistically before** the delete/repost
+  (so a retry fired mid-bump — or after the run dies before the final save — is
+  debounced and can't orphan a duplicate), `chat_delete`s the stored post,
+  `chat_postMessage`s the current `build_week_schedule` blocks, then
+  `save_schedule_post`s the new `ts`. If the old message is already gone, it
+  still reposts so the channel always ends with the schedule.
 
 ### Interaction with `_refresh_schedule_post` and the rollover
 

src/slack-bot/app.py

@@ -440,19 +440,27 @@ def _slack_error_code(exc) -> str | None:
         return None
 
 
-def handle_channel_message(event, client, schedule, schedule_channel):
+def handle_channel_message(event, client, schedule, schedule_channel, retry_num=None):
     """Bump the schedule post to the bottom of the channel on new activity.
 
     Debounced delete+repost of the tracked post so it stays at the bottom of
-    the channel timeline. Idempotent/safe: skips when there is no stored post,
-    when the triggering message is the bot's own / an edit-delete subtype, when
-    debounced inside the window, or when the bot isn't in the channel.
+    the channel timeline. Idempotent/safe: skips retried Slack deliveries
+    (``retry_num``), thread replies, the bot's own / edit-delete messages, when
+    there is no stored post, when debounced inside the window, or when the bot
+    isn't in the channel.
     """
+    # A retry means our first delivery already (likely) did the work but the
+    # 200 ack didn't reach Slack in time. Never bump again on a retry — the next
+    # real channel message bumps anyway — so a slow run can't double-post.
+    if retry_num:
+        logger.info("Ignoring retried message event (retry %s)", retry_num)
+        return
     if not schedule_channel or event.get("channel") != schedule_channel:
         return
     # Ignore the bot's own posts and non-user message events (edits, deletes,
-    # joins, …) — only genuine new user messages should bump.
-    if event.get("bot_id") or event.get("subtype"):
+    # joins, …), plus thread replies — a threaded reply doesn't push the
+    # schedule down the main timeline, so it isn't worth a delete+repost.
+    if event.get("bot_id") or event.get("subtype") or event.get("thread_ts"):
         return
 
     post = schedule.get_schedule_post(schedule_channel)
@@ -467,6 +475,16 @@ def handle_channel_message(event, client, schedule, schedule_channel):
     ):
         return
 
+    # Optimistically stamp the debounce window *before* the delete/repost, so a
+    # retry fired while this run is still in flight (or after it dies mid-bump)
+    # is suppressed and can't orphan a duplicate schedule message.
+    schedule.save_schedule_post(
+        schedule_channel,
+        post["message_ts"],
+        post.get("week_start", ""),
+        last_bump_ts=now,
+    )
+
     blocks = build_week_schedule(schedule)
     text = _schedule_fallback_text()
     try:
@@ -2199,7 +2217,10 @@ def create_app(
         publish_home(client, event["user"], _changelog_text())
 
     @app.event("message")
-    def handle_message_event(event, client):
-        handle_channel_message(event, client, schedule, schedule_channel)
+    def handle_message_event(event, client, request):
+        retry_num = request.headers.get("x-slack-retry-num") if request else None
+        handle_channel_message(
+            event, client, schedule, schedule_channel, retry_num=retry_num
+        )
 
     return app

tests/slack_bot/test_channel_bump.py

@@ -105,6 +105,56 @@ def test_skips_other_channel(slackbot_app, schedule, seed, client):
     client.chat_postMessage.assert_not_called()
 
 
+@freeze_time(MON)
+def test_retried_event_is_a_noop(slackbot_app, schedule, seed, client):
+    # A Slack retry (X-Slack-Retry-Num set) must never trigger a second bump.
+    seed.schedule_post(CHANNEL, "111.111")
+
+    slackbot_app.handle_channel_message(
+        _msg(), client, schedule, CHANNEL, retry_num="1"
+    )
+
+    client.chat_delete.assert_not_called()
+    client.chat_postMessage.assert_not_called()
+    # The record is left untouched (no optimistic debounce stamp either).
+    post = schedule.get_schedule_post(CHANNEL)
+    assert post["message_ts"] == "111.111"
+    assert post.get("last_bump_ts") is None
+
+
+@freeze_time(MON)
+def test_thread_reply_is_a_noop(slackbot_app, schedule, seed, client):
+    # A threaded reply doesn't push the schedule down the main timeline.
+    seed.schedule_post(CHANNEL, "111.111")
+
+    slackbot_app.handle_channel_message(
+        _msg(thread_ts="050.000"), client, schedule, CHANNEL
+    )
+
+    client.chat_delete.assert_not_called()
+    client.chat_postMessage.assert_not_called()
+    assert schedule.get_schedule_post(CHANNEL)["message_ts"] == "111.111"
+
+
+@freeze_time(MON)
+def test_stamps_debounce_before_repost(slackbot_app, schedule, seed, client):
+    # The debounce window is stamped optimistically before the delete/repost so
+    # a retry mid-bump is suppressed even if this run dies before the final save.
+    seed.schedule_post(CHANNEL, "111.111")
+    captured = {}
+
+    def _delete(**kwargs):
+        # Mid-bump snapshot: the record already carries last_bump_ts.
+        captured["post"] = schedule.get_schedule_post(CHANNEL)
+
+    client.chat_delete.side_effect = _delete
+    client.chat_postMessage.return_value = {"ts": "222.222"}
+
+    slackbot_app.handle_channel_message(_msg(), client, schedule, CHANNEL)
+
+    assert captured["post"].get("last_bump_ts") is not None
+
+
 @freeze_time(MON)
 def test_skips_when_bot_not_in_channel(slackbot_app, schedule, seed, client):
     seed.schedule_post(CHANNEL, "111.111")