Ferron crashes

[zdila]

RUST_BACKTRACE=1 /home/martin/ferron/target/debug/ferron -c sample.kdl

thread 'Request handler' (2174346) panicked at ferron/src/handler.rs:105:8:
called `Result::unwrap()` on an `Err` value: Os { code: 12, kind: OutOfMemory, message: "Cannot allocate memory" }
stack backtrace:
   0: __rustc::rust_begin_unwind
             at /rustc/ded5c06cf21d2b93bffd5d884aa6e96934ee4234/library/std/src/panicking.rs:698:5
   1: core::panicking::panic_fmt
             at /rustc/ded5c06cf21d2b93bffd5d884aa6e96934ee4234/library/core/src/panicking.rs:80:14
   2: core::result::unwrap_failed
             at /rustc/ded5c06cf21d2b93bffd5d884aa6e96934ee4234/library/core/src/result.rs:1862:5
   3: core::result::Result<T,E>::unwrap
             at /home/martin/.rustup/toolchains/stable-x86_64-unknown-linux-gnu/lib/rustlib/src/rust/library/core/src/result.rs:1233:23
   4: ferron::handler::create_http_handler::{{closure}}
             at /home/martin/ferron/ferron/src/handler.rs:105:8
Error while running a server: receiving from an empty and closed channel

(sample.kdl can be anything, even * {})

report-bba427e0-c918-4fb9-8271-908297daf424.toml.gz

Happens with version 2.3.2:

• ferron-2.3.2-i686-unknown-linux-gnu.zip
• ferron-2.3.2-i686-unknown-linux-musl.zip
• locally compiled in dev or release mode

Workaround:

Replace available_parallelism with 8 at

ferron/ferron/src/main.rs

Line 1425 in ac10f2a

for _ in 0..available_parallelism {

My PC has 24 CPU cores and 64GB RAM.

[DorianNiemiecSVRJS]

Thank you for reporting the issue!

I have successfully reproduced the bug on Ferron 2.3.2 for 32-bit Linux with musl libc (i686-unknown-linux-musl) target downloaded from https://dl.ferron.sh/.

I have run the ulimit -l 1024 command, and then started Ferron, and it crashed with this TOML-format report file:

name = "Ferron"
operating_system = "Linux (Arch Linux) [x86_64]"
crate_version = "2.3.2"
explanation = """
Panic occurred in file 'ferron/src/handler.rs' at line 105
"""
cause = 'called `Result::unwrap()` on an `Err` value: Os { code: 12, kind: OutOfMemory, message: "Out of memory" }'
method = "Panic"
backtrace = """
   0:  0x80cc350 - <unresolved>
   1:  0x8c74b5f - <unresolved>
   2:  0x8c982b2 - <unresolved>
   3:  0x8c98216 - <unresolved>
   4:  0x8c981f9 - <unresolved>
   5:  0x8066b27 - <unresolved>
   6:  0x80670b2 - <unresolved>
   7:  0x810fbaf - <unresolved>
   8:  0x8118b6c - <unresolved>
   9:  0x8ca1ed8 - <unresolved>
"""

It seems like this issue is related to io_uring (which is enabled by default in Ferron), see https://man7.org/linux/man-pages/man2/io_uring_register.2.html and ENOMEM error. The ulimit command I have used to reproduce the bug decreases the RLIMIT_MEMLOCK.

And from the man page for io_uring_register:

       ENOMEM Insufficient kernel resources are available, or the caller
              had a non-zero RLIMIT_MEMLOCK soft resource limit, but
              tried to lock more memory than the limit permitted.  This
              limit is not enforced if the process is privileged
              (CAP_IPC_LOCK).

Perhaps this needs better error reporting or handling (reporting the error instead of crashing)?

When running the web server with io_uring entirely disabled using a global io_uring #false directive, it starts just fine. So perhaps falling back to having io_uring disabled and logging a warning would work?

[zdila]

Maybe if io_uring directive is not set then Ferron could first try to start with io_uring enabled and if it fails then print a warning and start with io_uring disabled.

[DorianNiemiecSVRJS]

Maybe if io_uring directive is not set then Ferron could first try to start with io_uring enabled and if it fails then print a warning and start with io_uring disabled.

Thank you!

I have implemented fallback with io_uring disabled if io_uring couldn't be initialized (and the option to enable io_uring isn't explicitly specified) in the 7f166f4 commit, along with other commits. This change is also going to arrive in the next version of Ferron 2.