Clearly document the cases where fork() is supported in grpc()

[sreecha]

gRPC does not support fork() in general - i.e it is not safe for a gRPC application to call fork() as it can lead to undefined behavior.

As of release 1.7, we added very limited support to fork(); where it is safe for an application to call fork() as long as the application did not start any gRPC work (beyond calling grpc_init()) prior to calling fork()`

We need to do a better job at documenting this clearly - the fact that gRPC does not support fork() in general and the only time it is safe to call fork() is if the application did not start any gRPC work.

We have been noticing several issues caused due to fork() (and this is mainly due to grpc users not realizing that fork() is not supported)

[sreecha]

Reassigning to @kpayson64 .

[zhanghaohit]

I noticed a fork.h header. does it provide the support for fork?

void grpc_prefork();
void grpc_postfork_parent();
void grpc_postfork_child();

Thanks.

[srini100]

That is the intention but first this bug needs to be fixed to do fork/exec.

@zhanghaohit what is your use case?

[zhanghaohit]

My use case is:
I have to fork a new process, which will take over the job from the main process. It means, after the forking, the main process will not do anything with grpc.

But after the child process finished its job, it just exists. Then the main process will fork a new child process again.

So I have to fork multiple times from the main process.

@srini100 Can I achieve this in the current grpc implementation?

Do I have to destroy the Stub instance and re-create in the child process?

Thanks.

[srini100]

@zhanghaohit, forking with an active stub will put you in a situation where parent and child will share the same connection. gRPC currently doesn't cleanup FDs in child process. IMO, recreating the stub in child process should work.

[zhanghaohit]

@srini100 Thanks for your reply.

Yes. I did recreate the stub in the child process. In my case, do I need to call grpc_prefork(), grpc_postfork_parent(), grpc_postfork_child() also?

But anyway, no matter I call grpc_fork_ or not, it will hang there after second fork.

[srini100]

These functions are called internally by gRPC. You shouldn't call these.

[zhanghaohit]

Oh. ic. but still cannot get it to work by recreating the stub in the child process. It will hang after the second fork.

[kpayson64]

Closed via #14870