--safe-crash-log-file flag

[d-netto]

PR Description

Creates a --safe-crash-log-file that is used to redirect crash logs created while we were running the signal handler.

For now it's only x86_64 Linux but shouldn't be that hard to port it to MacOS.

Example usage:

./julia --safe-crash-log-file=/tmp/foobar.txt -e "p = Ptr{Any}(C_NULL); b = unsafe_load(p, 0)"

Which leads to:

cat /tmp/foobar.txt 

[83691] signal (11.1): Segmentation fault
in expression starting at none:1
signal (11) thread (1) unsafe_load at ./pointer.jl:119
signal (11) thread (1) unknown function (ip: 0x7f477566a17f)
signal (11) thread (1) _jl_invoke at /home/ubuntu/julia-RAI/src/gf.c:2895 [inlined]
signal (11) thread (1) ijl_apply_generic at /home/ubuntu/julia-RAI/src/gf.c:3077
signal (11) thread (1) jl_apply at /home/ubuntu/julia-RAI/src/julia.h:1982 [inlined]
signal (11) thread (1) do_call at /home/ubuntu/julia-RAI/src/interpreter.c:126
signal (11) thread (1) eval_value at /home/ubuntu/julia-RAI/src/interpreter.c:223
signal (11) thread (1) eval_stmt_value at /home/ubuntu/julia-RAI/src/interpreter.c:174 [inlined]
signal (11) thread (1) eval_body at /home/ubuntu/julia-RAI/src/interpreter.c:617
signal (11) thread (1) jl_interpret_toplevel_thunk at /home/ubuntu/julia-RAI/src/interpreter.c:775
signal (11) thread (1) jl_toplevel_eval_flex at /home/ubuntu/julia-RAI/src/toplevel.c:934
signal (11) thread (1) jl_toplevel_eval_flex at /home/ubuntu/julia-RAI/src/toplevel.c:877
signal (11) thread (1) jl_toplevel_eval_flex at /home/ubuntu/julia-RAI/src/toplevel.c:877
signal (11) thread (1) ijl_toplevel_eval_in at /home/ubuntu/julia-RAI/src/toplevel.c:985
signal (11) thread (1) eval at ./boot.jl:385 [inlined]
signal (11) thread (1) exec_options at ./client.jl:291
signal (11) thread (1) _start at ./client.jl:552
signal (11) thread (1) jfptr__start_83344 at /home/ubuntu/julia-RAI/usr/lib/julia/sys.so (unknown line)
signal (11) thread (1) _jl_invoke at /home/ubuntu/julia-RAI/src/gf.c:2895 [inlined]
signal (11) thread (1) ijl_apply_generic at /home/ubuntu/julia-RAI/src/gf.c:3077
signal (11) thread (1) jl_apply at /home/ubuntu/julia-RAI/src/julia.h:1982 [inlined]
signal (11) thread (1) true_main at /home/ubuntu/julia-RAI/src/jlapi.c:582
signal (11) thread (1) jl_repl_entrypoint at /home/ubuntu/julia-RAI/src/jlapi.c:731
signal (11) thread (1) main at /home/ubuntu/julia-RAI/cli/loader_exe.c:58
signal (11) thread (1) __libc_start_main at /lib/x86_64-linux-gnu/libc.so.6 (unknown line)
signal (11) thread (1) _start at ./julia (unknown line)
Allocations: 2908 (Pool: 2898; Big: 10); GC: 0

Checklist

Requirements for merging:

• I have opened an issue or PR upstream on JuliaLang/julia: NIL
• I have removed the port-to-* labels that don't apply.
• I have opened a PR on raicode to test these changes: NIL

[li1]

We have experimented with this on LATEST, changes LGTM through my Julia-noobish eyes. Would be great if @d-netto and/or @kpamnany can take a look at the changes we had to introduce to make this play nice with the OTel ingest.

[kpamnany]

jl_inside_signal_handler() will return 1 for the mutator threads when they're stopped for GC... do they ever call jl_safe_printf while they're waiting?

That's the only question I have about this... are there any other situations where we call jl_safe_printf in signal handler context besides when we're emitting a stack trace.

[adnan-alhomssi]

Comment for the future:
The signal handler is in

julia/src/signal-handling.c

Line 438 in ba5345f

{

void jl_critical_error(int sig, int si_code, bt_context_t *context, jl_task_t *ct)
which calls
https://github.com/RelationalAI/julia/blob/v1.10.2%2BRAI/src/stackwalk.c#L658 void jl_print_bt_entry_codeloc(int sig, jl_bt_element_t *bt_entry) JL_NOTSAFEPOINT

We could work up in the stack to emit a single JSON line that includes all backtrace lines if we want to.

[kpamnany]

I think this is okay. I don't much like the name --safe-crash-log-file, but no need for bike-shedding atm. :)