Attention is currently required from: neels, pespin.
1 comment:
Patchset:
Your responses on the blocking behavior still do not have empirical substance. […]
I am surprised to hear that the VTY would use blocking writes. telnet_new_connection() uses osmo_fd_register() on the accept()ed socket, which as it first step sets the Fd non-blocking. libosmovty's buffer_flush_available() at least claims to deal with non-blocking sockets.
Likewise, every new CTRL file descriptor goes through osmo_fd_register() which sets O_NONBLOCK as described above.
Which blocking "various syscalls" are you referring to? I would generally consider those bugs and it would be good to document those as such if you know any.
The general intent was always to have no blocking syscalls, specifically no blocking I/O from osmo* programs. I know of very few exceptions (like parts of osmo-e1d) and there it is a TODO to get them fixed and there are related comments in the code.
So the point is not how long nft takes [in your testing], or whether we trust you. The point is that doing anything non-blocking is - AFAICT - a grave violation of the general architecture and paradigm of how we do things in the non-blocking, asynchronous, event-driven osmo-* world.
To view, visit change 36540. To unsubscribe, or for help writing mail filters, visit settings.