Issue #19790 has been updated by matz (Yukihiro Matsumoto). I accept this proposal, at least we can experiment. I am a bit worried about calling open(2) in other functions/system calls from the signal handler. Matz. ---------------------------------------- Feature #19790: Optionally write Ruby crash reports into a file rather than STDERR https://bugs.ruby-lang.org/issues/19790#change-104268 * Author: byroot (Jean Boussier) * Status: Open * Priority: Normal ---------------------------------------- ### Use case On our servers we set [`/proc/sys/kernel/core_pattern`](https://man7.org/linux/man-pages/man5/core.5.html) to point to a small utility that report all the crashes happening in production with the associated core dump into BugSnag. This allowed us to find and fix many Ruby and native extensions bugs in the last few years. However, these are hard to triage, so we'd like to augment these crash reports with the output of `rb_vm_bugreport()`. ### Problem `rb_vm_bugreport()` is hard coded to print to STDERR, this makes it hard to extract and parse the report in a production environment, as very often STDERR is shared with other processes, so the crash report is intertwined with logs from other processes. ### Feature Request It would be very useful if Ruby could write the crash report to an arbitrary path rather than STDERR, akin to `kernel/core_pattern`. Especially it would be useful if it supported interpolating the crashing process PID with `%p` like `kernel/core_pattern`, as it would make it easier to map that report with the core file. This could be controller by an environment variable such as `RUBY_BUGREPORT_PATH`. e.g. ``` RUBY_BUGREPORT_PATH=/var/log/ruby/ruby-crash-pid-%p.log ``` ### Optional Features `kernel/core_pattern` supports other interpolations, however not all of them would make sense for Ruby to support. %% A single % character. %c Core file size soft resource limit of crashing process (since Linux 2.6.24). %d Dump mode—same as value returned by prctl(2) PR_GET_DUMPABLE (since Linux 3.7). %e The process or thread's comm value, which typically is the same as the executable filename (without path prefix, and truncated to a maximum of 15 characters), but may have been modified to be something different; see the discussion of /proc/pid/comm and /proc/pid/task/tid/comm in proc(5). %E Pathname of executable, with slashes ('/') replaced by exclamation marks ('!') (since Linux 3.0). %g Numeric real GID of dumped process. %h Hostname (same as nodename returned by uname(2)). %i TID of thread that triggered core dump, as seen in the PID namespace in which the thread resides (since Linux 3.18). %I TID of thread that triggered core dump, as seen in the initial PID namespace (since Linux 3.18). %p PID of dumped process, as seen in the PID namespace in which the process resides. %P PID of dumped process, as seen in the initial PID namespace (since Linux 3.12). %s Number of signal causing dump. %t Time of dump, expressed as seconds since the Epoch, 1970-01-01 00:00:00 +0000 (UTC). %u Numeric real UID of dumped process. Additionally, if `kernel/core_pattern` starts with a pipe (`|`), then it allows to pipe the core dump to another program, this may also make sense as a feature. ### Prior Art Aside from `kernel/core_pattern`, some other virtual machine have a similar feature, for instance the JVM has a configurable crash report: ``` -XX:ErrorFile=/var/log/java/hs_err_pid%p.log ``` -- https://bugs.ruby-lang.org/