System V IPC is missing

[fehrin]

When starting gvim i get
(gvim:2390): Gdk-WARNING **: shmget failed: error 38 (Function not implemented)
It seems that while Posix shared memory is present, System V IPC is still missing.

[russalex]

Just to avoid any ambiguity, my comments on #61 are about the filesystem perf delta. We know there are issues there and are looking into it.

IPC is on the backlog but it is not showing up in our telemetry very often so not terribly high. That said, PostgreSQL is something we want to start testing with so it may bubble up the priority stack as we get to that point.

[fpqc]

@russalex Are you doing telemetry on Azure or on Win10 Insiders (or both?) to decide this kind of priority?

[russalex]

We look mostly at telemetry from Insiders and a little at the Anniversary Update. We are trying to keep our finger on what the community is trying to run and the crashes / unimplemented syscalls which are blocking scenarios.

For items like PostgreSQL we generally look at comments here, input from folks within MSFT who work with developers, tweets, and blog / forum posts in places like stack overflow. We do talk with Azure on occasion but do not have a formal channel at this time.

More than happy to go a bit more into detail on our telemetry. It may make for a good blog post at some point.

[chesterm8]

If java uses the System V IPC to fork processes (which seems to be the root cause of #851) then this could be quite a significant issue.

[therealkenc]

No sorry to lead you astray on that. It's cloning with CLONE_SYSVSEM, but that's not what is causing the failure at the moment. That clone succeeds (message). It just looked like a red flag at the time I wrote the message, but you're dying on a different syscall.

[QinHuasong]

@therealkenc what should I do to solve the problem I mentioned in #1443 ?

[therealkenc]

There isn't a lot you can do at the moment because the syscall surface just isn't there yet. If it's your own code you might consider using a different IPC mechanism for message queues. The reason SysV IPC isn't getting a lot of telemetry hits is because it just isn't used that much anymore (for some arbitrary value of "that much").

[QinHuasong]

@therealkenc Thanks for your answer. I've tried POSIX mq instead of sysV msgget, but error message still there. It says:
mq_open error(38 - Function not implemented).
where I print out the errno and error msg.
Any other suggestion?

[therealkenc]

Apologies; I should have been more clear. I meant use AF_NETLINK or AF_UNIX. [Why have one way to send some bytes between process domains when you can have half a dozen, right?] It's not as bad as it sounds if you already have your IPC abstracted somewhat in your code.

[benhillis]

System V shared memory and semaphores were implemented in Creators Update. Message queues are on the backlog.

[ghost]

The bug is thus closed benhillis?

I tried to start gvim, it works fine (I am using xming for the xorg-part on win10, which seems to work fine for most programs; I have a problem with kde konsole but I suppose I may file another issue eventually; It's only some days since I even started with win10 altogether, including the *nix subsystem).

[benhillis]

Closing this, message queues are tracked by #1443.