Horton

Re: Optimize LISTEN/NOTIFY

Chao Li <li.evan.chao@gmail.com>

View thread
From: Chao Li <li.evan.chao@gmail.com>
To: Joel Jacobson <joel@compiler.org>
Cc: Tom Lane <tgl@sss.pgh.pa.us>, pgsql-hackers <pgsql-hackers@postgresql.org>
Date: 2025-10-16T02:54:12Z
Lists: pgsql-hackers

Commits

Same data as JSON: GET /api/v1/messages/:b64id/commits the thread's linked commits as JSON, with link sources. API reference →

  1. Optimize LISTEN/NOTIFY via shared channel map and direct advancement.

  2. Fix incorrect logic for caching ResultRelInfos for triggers


> On Oct 15, 2025, at 23:36, Joel Jacobson <joel@compiler.org> wrote:
> 
>> I agree with Tom that GetPendingNotifyChannels() is too heavy and unnecessary.
>> 
>> In PreCommit_Notify(), we can maintain a local hash table to record 
>> pending nofications’ channel names. dahash also supports hash table in 
>> local memory.
> 
> I'm confused, I assume you mean "dynahash" since there is no "dahash"
> in the sources? I see dynahash has local-to-a-backend support,
> but I don't see why we would need a hash table for this,
> we just iterate over it once in SignalBackends,
> I think the local list is fine.
> 
> The latest version gets rid of GetPendingNotifyChannels()
> and replaces it with the local list pendingNotifyChannels.

Sorry for the typo, Yes, I meant to dynahash” that you have already been using it.

In v18, I see you are building “pendingNotifyChannels” in PreCommit_Notify() with “List”:

```
+		/*
+		 * Build list of unique channels for SignalBackends().
+		 */
+		pendingNotifyChannels = NIL;
+		foreach_ptr(Notification, n, pendingNotifies->events)
+		{
+			char	   *channel = n->data;
+
+			/* Add if not already in list */
+			if (!list_member_ptr(pendingNotifyChannels, channel))
+				pendingNotifyChannels = lappend(pendingNotifyChannels, channel);
+		}
```

My suggestion of using dynahah was for the same purpose. Because list_member_ptr() iterates through all list nodes until find the target, so this code is still O(n^2).

Using a hash will make it faster. I used to work on project Concourse [1]. The system is heavily using the LISTEN/NOTIFY mechanism. There would be thousands of channels at runtime. In that case, hash search would be much faster than linear search.

[1] https://github.com/concourse/concourse

Best regards,
--
Chao Li (Evan)
HighGo Software Co., Ltd.
https://www.highgo.com/