Skip to content

Duplicate mmap log file at the beginning of darshan_core_shutdown #1053

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
wants to merge 7 commits into
base: main
Choose a base branch
from
Open

Duplicate mmap log file at the beginning of darshan_core_shutdown #1053

Show file tree
Hide file tree
Changes from 4 commits
Commits
Show all changes
7 commits
Select commit Hold shift + click to select a range
dad0e9b
Duplicate mmap log file at the beginning of darshan_core_shutdown
wkliao Jul 15, 2025
e081fe5
Move calls to unlink up, to right after darshan_log_finalize()
wkliao Jul 18, 2025
99f568b
printf on screen whether or not configure option --enable-mmap-logs i…
wkliao Jul 18, 2025
7af3cb4
reuse mmap file size, mmap_size, added as a member of darshan_core_ru…
wkliao Jul 22, 2025
14a755c
silence gcc 14.2 warning by checking return values
wkliao Aug 21, 2025
6a5c60f
fix snprintf length argument
wkliao Aug 21, 2025
56f0c82
Use ".dup.darshan" suffix for duplicated log file name
wkliao Aug 22, 2025
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
1 change: 1 addition & 0 deletions darshan-runtime/configure.ac
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -932,6 +932,7 @@ if test "x$enable_darshan_runtime" = xyes ; then
MDHIM module support - $enable_mdhim_mod
HEATMAP module support - $enable_heatmap_mod
LDMS runtime module - $enable_ldms_mod
Use mmap to log file - $enable_mmap_logs
Memory alignment in bytes - $with_mem_align
Log file env variables - $__log_path_by_env
Location of Darshan log files - $__log_path
Expand Down
51 changes: 45 additions & 6 deletions darshan-runtime/lib/darshan-core.c
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -433,6 +433,10 @@ void darshan_core_shutdown(int write_log)
darshan_record_id *mod_shared_recs = NULL;
int shared_rec_cnt = 0;
#endif
#ifdef __DARSHAN_ENABLE_MMAP_LOGS
char dup_log_fame[__DARSHAN_PATH_MAX + 5];
dup_log_fame[0] = '\0';
#endif

/* disable darhan-core while we shutdown */
__DARSHAN_CORE_LOCK();
Expand Down Expand Up @@ -470,13 +474,34 @@ void darshan_core_shutdown(int write_log)
internal_timing_flag = final_core->config.internal_timing_flag;

#ifdef __DARSHAN_ENABLE_MMAP_LOGS
/* remove the temporary mmap log files */
/* NOTE: this unlink is not immediate as it must wait for the mapping
* to no longer be referenced, which in our case happens when the
* executable exits. If the application terminates mid-shutdown, then
* there will be no mmap files and no final log file.
/* Flush memory-mapped data to the underlying file and then duplicate the
* mmap log file, in case an interrupt happens before the completion of
* this subroutine, leaving the log file corrupted. See github issue #1052.
*/
unlink(final_core->mmap_log_name);
msync(final_core->log_hdr_p, final_core->mmap_size, MS_SYNC);

/* Duplicate the log file, Below we on purpose ignore errors from open(),
* lseek(), read(), write(), and close(), as they are not fatal and should
* not affect the remaining tasks of this subroutine, i.e. log data file
* processing. Note if any of these system calls failed, then the contents
* of duplicated file will become useless, but such problem will be even
* more serious than being unable to duplicate the file.
*/
int mmap_fd = open(final_core->mmap_log_name, O_RDONLY, 0644);
if (mmap_fd != -1) {
int dup_mmap_fd;
void *buf = (void*) malloc(final_core->mmap_size);
read(mmap_fd, buf, final_core->mmap_size);
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

FYI, this block of code produces the following warning with gcc 14.2:

../../../darshan-runtime/lib/darshan-core.c: In function ‘darshan_core_shutdown’:
../../../darshan-runtime/lib/darshan-core.c:494:9: warning: ignoring return value of ‘read’ declared with attribute ‘warn_unused_result’ [-Wunused-result]
  494 |         read(mmap_fd, buf, final_core->mmap_size);
      |         ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
../../../darshan-runtime/lib/darshan-core.c:500:13: warning: ignoring return value of ‘write’ declared with attribute ‘warn_unused_result’ [-Wunused-result]
  500 |             write(dup_mmap_fd, buf, final_core->mmap_size);
      |             ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

Copy link
Collaborator Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I ignored them on purpose, as these codes are for file duplication, wouldn't
want any error stop the duplication.

Anyway, I added checks in 14a755c.
Strangely, I also run gcc 14.2.1, but did not get those warnings.
Please see if they are silenced.

close(mmap_fd);
snprintf(dup_log_fame, strlen(final_core->mmap_log_name), "%s.dup",
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

This snprintf truncates the files name to something like /tmp/carns_python3_id1445957_mmap-log-3779006568372861245-0.darsha because the 2nd argument to is calling strlen() on the original mmap_log_name. Maybe this size limit on the snprintf should be based on the size of the dup_log_fname instead?

Copy link
Collaborator Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

This should be fixed in 6a5c60f.

Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Actually I have one other request here. Sorry this didn't occur to me sooner. I like that the duplicate file has "dup" in the file name, but it may be better if it weren't used as the file extension. I find myself doing things like ls /tmp/*.darshan to look for intermediate files, when with this patch it would really need to be ls /tmp/*.darshan /tmp/*.darshan.dup to find them all.

Can we change the naming convention so that even the duplicate file names end with .darshan?

Copy link
Collaborator Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Another way is to run "ls /tmp/*.darshan*".

The log file name is constructed in:

darshan/darshan-runtime/lib/darshan-core.c

Lines 830 to 832 in b7e4e89

snprintf(core->mmap_log_name, __DARSHAN_PATH_MAX,
"/%s/%s_%s_id%d_mmap-log-%" PRIu64 "-%d.darshan",
core->config.mmap_log_path, cuser, __progname, jobid, logmod, my_rank);

Where do you suggest to insert substring "dup"?

Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

There can be other things in /tmp with "darshan" in the name that aren't log files, though (I have a bunch for sure, though I'm an outlier because I've been experimenting with a lot of Darshan things :))

I'm not picky about where to put dup in the name (other than not at the end). Maybe just making the extension .dup.darshan?

Copy link
Collaborator Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Updated. Please give it a try.

final_core->mmap_log_name);
dup_mmap_fd = open(dup_log_fame, O_CREAT | O_WRONLY, 0644);
if (dup_mmap_fd != -1) {
write(dup_mmap_fd, buf, final_core->mmap_size);
close(dup_mmap_fd);
}
free(buf);
}
#endif

final_core->comp_buf = malloc(final_core->config.mod_mem);
Expand Down Expand Up @@ -692,6 +717,15 @@ void darshan_core_shutdown(int write_log)
/* finalize log file name and permissions */
darshan_log_finalize(logfile_name, start_log_time);

#ifdef __DARSHAN_ENABLE_MMAP_LOGS
/* Now, we are done with accessing the temporary log files and thus can
* safely remove them. We ignore the error code returned from unlink(), as
* there is no harm to unlink a non-existing file.
*/
unlink(final_core->mmap_log_name);
if (dup_log_fame[0] != '\0') unlink(dup_log_fame);
#endif

if(internal_timing_flag)
{
double open_tm;
Expand Down Expand Up @@ -770,14 +804,17 @@ void darshan_core_shutdown(int write_log)
for(i = 0; i < DARSHAN_KNOWN_MODULE_COUNT; i++)
if(final_core->mod_array[i])
final_core->mod_array[i]->mod_funcs.mod_cleanup_func();

darshan_core_cleanup(final_core);

#ifdef HAVE_MPI
if(using_mpi)
{
free(shared_recs);
free(mod_shared_recs);
}
#endif

free(logfile_name);

return;
Expand Down Expand Up @@ -806,6 +843,8 @@ static void *darshan_init_mmap_log(struct darshan_core_runtime* core, int jobid)
if(mmap_size % sys_page_size)
mmap_size = ((mmap_size / sys_page_size) + 1) * sys_page_size;

core->mmap_size = mmap_size;

darshan_get_user_name(cuser);

/* generate a random number to help differentiate the temporary log */
Expand Down
1 change: 1 addition & 0 deletions darshan-runtime/lib/darshan.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -144,6 +144,7 @@ struct darshan_core_runtime
char *comp_buf;
#ifdef __DARSHAN_ENABLE_MMAP_LOGS
char mmap_log_name[__DARSHAN_PATH_MAX];
size_t mmap_size;
#endif
#ifdef HAVE_MPI
MPI_Comm mpi_comm;
Expand Down