A running description of activity related to DragonFly BSD.
So much that I’m doing bullet points:
In addition to committing the new scheduler improvements mentioned earlier this week, Matthew Dillon has made some changes to how DragonFly handles low memory situations, so the system will be able to recover much more quickly. He’s also asking for testers of his new vm.burst_fault sysctl.
Sepherosa Ziehau’s recent change to libpcap means that dumps of network traffic taken before this change won’t be readable with libpcap after this change. Unlikely to affect anyone unless you are both dumping a lot of data off the network and updating your system rapidly, but worth mentioning.
I don’t know if this is going to be the long-term solution, as discussion is ongoing, but the existing commit mail format has been explained.
Hasso Tepper has a copy of cgit running for DragonFly’s git repository, and I’ve set up the same thing on leaf.
When creating the 2.1 tag in git, Simon ‘corecode’ Schubert also auto-created a message showing every commit ever, grouped by committer, with the first line of each commit. Reading through it provides an interesting look at what particular itch different people have scratched with DragonFly, over the years. (Also available as a plain count.)
Matthew Dillon’s updated Hammer to create two new directives: ‘version’ and ‘version-upgrade’, along with a number of internal changes and fixes.
The new switch to git has brought out a lot of new committer activity; nothing to point you at specifically, but it’s nice to see that it has encouraged action.
Michael Neumann has come up with a way to automatically create pseudo file systems (PFS) when mirroring a Hammer volume. Previously, the destination/slave file system would have to be created first; this makes it Just Work.
This means Hammer data streams will be incompatible with versions before and after this change, but it’s not going to damage anything. Introducing a versioning system into Hammer data streams is an available project…
Matthew Dillon has made a scheduler change that apparently improves responsiveness when CPU load is high, to fix an issue reported by Simon ‘corecode’ Schubert. Please test if you are running bleeding-edge code.
Hasso Tepper has synced sensorsd(8), the sensor framework in DragonFly, with the latest version in OpenBSD.
Hasso Tepper has added (based on this FreeBSD work) the ability to “parse the utrace(2) entries generated by malloc(3) in a more human-readable format”.
Sascha Wildner is updating DragonFly to tzcode2008g, which will modernize our time system, along with making 64-bit time_t possible. It also apparently fixes a recently reported problem in Python. Sascha links to this time page in his message, with more time zone link information than ever I’ve seen.
Oh, and Sascha updated timezone data, too.
Hasso Tepper brought in a fix from OpenBSD for ssh; apparently empty banners on some types of network equipment would cause a disconnect. This isn’t major, but there may just be someone out there reading this for whom knowing about that saves a lot of frustration.
Sepherosa Ziehau has added code to make it possible to run network threads without the Big Giant Lock. It’s still experimental, so it has to be manually set via sysctl.
If you are running bleeding edge DragonFly, and you don’t mind panicing your system, Sepherosa Ziehau has made some changes. Specifically, if you see messages on your console about rtfree_remote(), set net.route.remote_free_panic to 1 and post a link to the resulting coredump.
Matthew Dillon made some changes to the scheduler; his commit message has some interesting details.
This sentence caught my eye from a recent commit by Sepherosa Ziehau: “it accidentally doubles the current lo0 performance“