Re: Maximum amount of uprobes and uprobe and uprobe_ret relation

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 



On Thu, Sep 19, 2024 at 5:13 PM Sebastião Amaro <sebassamaro97@xxxxxxxxx> wrote:
>
> Hi everyone!
> I have two questions related to user function probes:
> Firstly, I am trying to have a process attach more than 1024 uprobes,
> however, I am getting the error: "failed to create BPF link for
> perf_event FD 1023: -24 (Too many open files)" even after changing
> ulimit -n to 4096  github issue[1].

See [0], it might be that `ulimit -n` isn't really changing the limit
for your process or something. To be 100% sure I'd do
setrlimit(RLIMIT_NOFILE, ...) from inside the process to verify.

  [0] https://unix.stackexchange.com/questions/8945/how-can-i-increase-open-files-limit-for-all-processes

> Secondly, I am running some tests with uprobe and uprobe_ret in multiple
> functions in the redis binary, but I am noticing that when counting the
> times the uprobes and uprobes_ret are called, in the end they do not
> match 1 to 1. Either individually (a uprobe/uprobe_ret in the same
> function), or the total sum. Is this a predictable behaviour?
> I am tracing several functions in such as [2].

Attachment is not atomic, so you might get some uprobes attached
before corresponding uretprobe is attached, and vice versa. So counts
might not match 1:1 during attachment and detachment.

>
> [1]https://github.com/libbpf/libbpf-rs/issues/942
> [2]https://github.com/redis/redis/blob/3a3cacfefabf8ced79b448169319ce49cca2bfb7/src/rdb.c#L1782
>
> Thank you, and Best Regards,
> Sebastião Amaro
>





[Index of Archives]     [Linux Samsung SoC]     [Linux Rockchip SoC]     [Linux Actions SoC]     [Linux for Synopsys ARC Processors]     [Linux NFS]     [Linux NILFS]     [Linux USB Devel]     [Video for Linux]     [Linux Audio Users]     [Yosemite News]     [Linux Kernel]     [Linux SCSI]


  Powered by Linux