Asterisk 24: the new LTS, what it brings and when to upgrade

Asterisk 24: the new LTS, what it brings and when to upgrade

Asterisk 24.0.0 was released on November 5th. It is the new LTS version, taking over from 22 as the long-term reference. According to the official dates, it will have full support until October 15th, 2030 and security fixes until October 2031, so it is the version many of us will end up building PBXs and platforms on for the next few years. The question is not whether to migrate, but when.

An LTS that arrives without noise

The first thing you notice when you open the changelog is how discreet it is: 43 commits, 22 authors, 32 issues resolved and no security advisories. It is not a release of big headlines, and that makes sense. Since it is cut from master, 24 inherits everything accumulated since 22, including what 23 introduced (the outbound WebSocket channel, the removal of users.conf, DeadAGI and other deprecated features), and in an LTS the focus shifts to stabilizing rather than adding.

What’s new

chan_websocket, unbuffered mode. It is enabled with the u option of Dial() and forwards frames to the core as they arrive. It is designed for applications that already send framed audio with its own timing, but occasionally need to send audio in bulk with START_MEDIA_BUFFERING. There is also a new write_timeout parameter (100 ms by default) in chan_websocket.conf and websocket_client.conf, so that a server that stops reading does not end up blocking the channel in cascade. And a bug is fixed where the frame queue was never drained, turning any delay into permanent latency. For those of us building voice agents over WebSocket, this is the most relevant part of this release.

PJSIP and DNS. Qualify now probes all resolved destinations (SRV, A and AAAA) and marks the contact as reachable as soon as one of them answers. Before, a single destination was chosen, and a contact with other live destinations could end up marked as down.

ConfBridge. The new announcements=no option in the bridge profile avoids creating the announcer channel, saving one channel and one thread per conference.

Prometheus. The channels_detail_metrics_enabled and bridges_detail_metrics_enabled options let you stop exporting the detailed channel and bridge metrics, which generate a huge volume on high-traffic systems.

ARI. You can now record a channel that is being dialed with channels.dial; before, it failed with “Cannot record channel while in bridge”.

Call quality. The MES (Media Experience Score) calculation in res_rtp_asterisk is fixed: latency used the full RTT instead of the one-way delay, accumulated loss dragged the score down, and jitter was not scaled. If you alert on MES, review your thresholds, because scores will change compared to previous versions.

Stability. Four extension state bugs (hints and BLF) that appeared in 22.11.0, including a crash when changing a hint with a presence part and park slots that stayed in Unavailable; the device state of chan_dahdi (Polycom BLFs ignored a RINGING that came after an INUSE); ChanSpy delaying the Hangup of the spied channel; a use-after-free in func_speex; and a crash when evaluating CHANNEL(channeltype) on a dummy channel.

Watch out when upgrading

There is one detail worth reading before touching anything: on MySQL and MariaDB, the time column of queue_log changes from DATETIME to DATETIME(6) to keep the microseconds that Asterisk was already writing. The ALTER rewrites the table and, on a large table, blocks writes while it runs, so it is better to do it in a maintenance window. Alembic scripts are also added for res_geolocation. And if you are coming from 20 or earlier, remember that chan_sip disappeared in 21: any Dial(SIP/...) has to move to PJSIP, in addition to the removals that came with 23.

Is it worth upgrading now?

It depends on where you are. If you are still on 20 or earlier, yes: start planning it now, because the migration work (especially if you still use chan_sip) will not do itself, and 20 runs out of support in October 2027. If you are on 22 and everything works, there is no rush: you have two years of full support left, and the sensible thing is to test 24 in a lab and wait for the first point releases before touching production.

If I had to build a PBX today

This would be my approach. If the PBX has to go into production this month, I would install the 22 LTS with the latest point release, which is the most battle-tested option. If it is going to live for five years or more and I have a couple of months of margin, I would build on 24, but I would put it into production with 24.1 or 24.2. It is not distrust: a .0 is the first version to see real traffic, and the changelog itself illustrates it, with four extension state bugs from 22.11.0 being fixed right before the release. Asterisk ships point releases every six to eight weeks, so the wait is short.

More information and downloads at asterisk.org, and the full changelog for 24.0.0.

Your experience matters

Your experience matters

The best articles are born from good discussions.

If you have a Sinologic account, you don't need to fill in these fields. Sign up free · Log in

Open community

Join the Sinologic community

Create your free account and join the conversations about VoIP, Asterisk, Kamailio and IP telephony.

⭐
Verified name Your name appears with a member badge on every comment.
🔔
Notifications Get notified when someone replies to your comments.
👍
Reactions React with emojis to other users' comments.
👤
Customizable profile Avatar, bio, links to your Twitter, GitHub and Telegram.
📬
Exclusive newsletterComing soon Technical content and news straight to your inbox.
🧪
Early accessComing soon Test tools and features before anyone else.
Create free account