From 2a73585dfd5d22701e1f525ae5ecff60ef4813bd Mon Sep 17 00:00:00 2001
From: io mintz <io@mintz.cc>
Date: Thu, 8 Oct 2020 02:48:25 +0000
Subject: [PATCH] cluster startup no longer depend on the first cluster running

---
 cogs/stats.py | 39 ++++++++++++++++++++++++++++++++++-----
 1 file changed, 34 insertions(+), 5 deletions(-)

diff --git a/cogs/stats.py b/cogs/stats.py
index 02a063b..efdfa45 100644
--- a/cogs/stats.py
+++ b/cogs/stats.py
@@ -18,20 +18,49 @@ from multiprocessing.shared_memory import ShareableList
 from discord.ext import commands
 from bot_bin.stats import BotBinStats
 
+# The resource tracker unlinks shared memory on shutdown and assumes all processes that share memory
+# share a parent python process. But this is not necessarily the case: the user might decide to run
+# each cluster as a systemd unit for example. We don't want a crash of a single cluster to unlink
+# the shared memory segment, so we stub out the resource tracking process here,
+# as there's no built in way to disable it.
+# See https://bugs.python.org/issue38119
+from multiprocessing import resource_tracker
+resource_tracker.ensure_running = lambda: None
+resource_tracker.main = lambda: None
+
 class Stats(BotBinStats):
 	def __init__(self, bot):
 		super().__init__(bot)
-		seq = [0] * self.bot.shard_count if self.is_opener() else None
-		# Use our user ID as part of the shm name to allow running multiple instances of the bot on the same machine.
-		self.shlist = ShareableList(seq, name=f'emote-manager-{self.bot.user_id}')
+
+		# Use our user ID as part of the shm name
+		# to allow running multiple instances of the bot on the same machine.
+		shm_name = f'emote-manager-{self.bot.user_id}'
+		if self.is_opener():
+			seq = [0] * self.bot.shard_count
+			try:
+				self.shlist = ShareableList(seq, name=shm_name)
+			except FileExistsError:
+				# The file was probably left open from a previous run of the bot,
+				# so let's re-attach and wipe the old data.
+				self.shlist = ShareableList(name=shm_name)
+				# apparently ShareableLists don't support slicing with None values of `stop`
+				self.shlist[:self.bot.shard_count] = seq
+		else:
+			try:
+				self.shlist = ShareableList(name=shm_name)
+			except FileNotFoundError:
+				# looks like we're the first cluster to start up
+				seq = [0] * self.bot.shard_count
+				self.shlist = ShareableList(seq, name=shm_name)
+
 		self.count()
 
 	def is_opener(self):
-		"""return whether this is the process that should open the shared memory"""
+		"""return whether this is the cluster that should open the shared memory"""
 		return 0 in self.bot.shard_ids
 
 	def is_reporter(self):
-		"""return whether we should report stats to the bot lists"""
+		"""return whether this is the cluster that should report stats to the bot lists"""
 		return self.bot.shard_count - 1 in self.bot.shard_ids
 
 	def cog_unload(self):