-
Notifications
You must be signed in to change notification settings - Fork 0
/
README.new
63 lines (63 loc) · 4.76 KB
/
README.new
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
XALT 2.7+
(2.7.1) * Added contrib/TACC/build_default_2_account_str.in to build a json table that
maps users to default charge string.
* xalt_syslog_to_db.in.py and xalt_file_to_db.in.py support --u2acct that allows sites
to specify a default charge account string when it is not specify in the job script.
* xalt_usage_report.in.py support --queue to only report from a particular queue.
* PKG intermediate results are now written to /dev/shm/XALT_... instead of /tmp/XALT_...
by default.
(2.7.2) * Support for env var XALT_FILE_PREFIX.
* Make job_id filtering optional.
(2.7.3) * Use uuid hash value to chose the 000 to 996 directory when using XALT_FILE_PREFIX.
(2.7.4) * Sample all executables (MPI included) with MPI tasks counts < MPI_ALWAYS_RECORD
(2.7.5) * Update xalt_configuration_report about XALT_SAMPLING and MPI_ALWAYS_RECORD
(2.7.6) * Fix bugs introduced in version 2.7.5 about XALT_SAMPLING.o
* Better reporting of when myinit() and myfini() have completed their work.
* Move uuid generation from libxalt_init.so to xalt_run_submission
(2.7.7) * Use dlopen/dlsym to dynamically link-in the uuid and crypto libraries.
This allows XALT to be used with containers where libuuid.so and libcrypto.so
might not be in the container.
(2.7.8) * Remember to use xalt-2.7.8 as tag not 2.7.8
(2.7.9) * Fix bug introduced when adding --version to xalt_configuration_report.C
(2.7.10) * Stopping building libuuid. Instead copy libuuid.so from system to XALT's lib64 directory.
(2.7.11) * Allow yes0 and run0 as values for XALT_TRACING to only track proc=0.
* for TACC_config.py make 0 < t < 30 mins be 0.01% and 30mins < t < 120mins be 1%
(2.7.12) * Remove trailing newline from the shared libraries captured from /proc/$PID/maps.
* Ignore any trailing newlines and spaces in shared libraries when ingesting into DB.
(2.7.13) * Issue #19: To avoid a name collision between ambertools having their own function
named connect and libc's connect, XALT moved the call to xalt_syshost() from
xalt_initialize.c to the xalt_run_submission executable.
(2.7.14) * Fix bug that was created with 2.7.11. Now XALT_TRACING=yes works with all mpi tasks and
XALT_TRACING=yes0 just traces from mpi task 0.
(2.7.15) * Switch XALTdb.py and Rmap_XALT.py to use xalt_rmapT.json instead of jsonReverseMapT.json
* Update documentation to explain reverseMap file and function tracking
* ld.in only does function tracking if reverseMap file exists.
* use xalt_rmap_exists command to determine if reverseMap file exists.
(2.7.16) * To protect sites that use jsonReverseMapT.json for directory to module name mapping, I have
modified Rmap_XALT.py to look first for xalt_rmapT.json. If not found then search for
jsonReverseMapT.json and convert it to xalt_rmapT style.
(2.7.17) * Issue #20: Added a configure/Env. Var to control whether the complete command line
will be recorded.
(2.7.18) * Issue #21: README.md doc patch.
(2.7.19) * Fix bug with reporting GPU tracking when HAVE_DCGM=no and HAVE_NVML=yes in
xalt_configuration_report
(2.7.20) * Issue #22: Allow for the curl transmission style. Copy system libcurl into XALT
lib64 install directory and Rpath that directory into xalt_run_submission etc.
* Issue #23: Running XALT's "ld" alone now produces an error: No input files just
like the real ld.
(2.7.21) * Allow XALT to have separate sampling intervals for scalar or mpi programs.
(2.7.22) * Correctly handle the case where a static parent program is XALT tracking and
this parent program fork-exec's a non-static child program also under XALT
tracking.
(2.7.23) * Set the pid_str variable as late as possible.
(2.7.24) * Add nodename to pid_str. Free pid_str after use.
(2.7.25) * Zero memory locations before freeing data. Try not to put non-zero memory that
user code will find with malloc().
* issue #24: check that /sys/module/nvidia exists before trying to track the GPU's
* try to dlopen("libnvidia-ml.so.1") if libnvidia-ml.so cannot be found.
(2.7.26) * Have myfini() force xalt_run_submission to return the uuid to make sure that
xalt_run_submission finishes before the myfini() routine completes.
(2.7.27) * Support new table xalt_field_of_science which connects account string to
NSF's field_of_science and NSF Directorate
(2.7.28) * Adding SIG_USR2 to list of captured signals
* Correctly always return the UUID from myfini() in xalt_initialize.c