How to debug a never-returning open()?

[matthiasbeyer]

Hi,

I have an call to open() that does not return.

My code:

        trace!("Reading to buffer: {}", p.display());
        let mut buf = vec![];
        let mut file = tokio::fs::OpenOptions::new()
            .create(false)
            .create_new(false)
            .write(false)
            .read(true)
            .open(&p)
            .await?;

        trace!("Opening worked...");
        file.read_to_end(&mut buf)
            .await
            .with_context(|| anyhow!("Reading file: {}", p.display()))?;
        trace!("Reading worked...");

I get the log output from "Reading to buffer" with the path, the path exists on my FS.

strace reports this:

statx(AT_FDCWD, "/tmp/REDACTED", AT_STATX_SYNC_AS_STAT, STATX_ALL, {stx_mask=STATX_BASIC_STATS, stx_attributes=0, stx_mode=S_IFREG|0644, stx_size=5, ...}) = 0
write(2, "\33[0m\33[38;5;8m[\33[0m2020-12-09T13:"..., 189) = 189
write(2, "\33[0m\33[38;5;8m[\33[0m2020-12-09T13:"..., 190) = 190
futex(0x56166ce20df8, FUTEX_WAKE_PRIVATE, 1) = 1
futex(0x56166ce20da0, FUTEX_WAKE_PRIVATE, 1) = 1
futex(0x56166ce1fa88, FUTEX_WAIT_PRIVATE, 4294967295, NULL)

and blocks there (I waited up to 1:30 min). The file is less than 10 bytes.

How to debug this?

[matthiasbeyer]

(btw I also tried tokio::fs::read(&p), but that has the exact same issue... so I tried the above code instead...

[Darksonn]

It sounds like your runtime is configured with max_threads equal to core_threads. When this happens, the runtime is not able to execute spawn_blocking calls since there are no threads left for blocking operations.

[matthiasbeyer]

That doesn't add up to me.
Because wouldn't that mean that other stuff would still be running? If that was the case, the strace output would still yield lines, wouldn't it?
Also, I have a i7 with 4+4 cores... and the quoted lines are the only future that runs... so I doubt that there is more than one (blocking) future at that time...

[Darksonn]

Well async code can still run. How did you set up your runtime?

[matthiasbeyer]

#[tokio::main]
async fn main() -> Result<()>

Is that what you mean?

[Darksonn]

I guess that's not the reason then. When I run your code locally, it works fine.