| Commit message (Collapse) | Author | Age | Files | Lines |
... | |
|
|
|
|
|
|
|
|
|
|
| |
1) Fix for assertion failure mentioned in my last checkin. Thanks to
Quincey for the fix.
2) Added tools/h5recover/trecover_verifier.c -- forgot to do this
in the last checkin.
Tested on Phoenix (serial), Linew (serial) and Jam (parallel).
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
1) In H5F.c, modified several metadata cache related routines to
talk to the new cache, not the pre-journaling cache.
2) Enabled the API smoke check in cache2_journal.
3) Wrote a example of use of the journaling API and included
it as a test in cache2_journal.c. Some re-factoring of the
cache2 test code to move supporting macros and functions
out of cache2_api.c and into cache2_common.c & .h.
4) Modified tools/h5recover/trecover to include a new verify
option, which is intended to verify that a file has been
correctly recovered via h5recover.
The basic idea of this feature is to look at the data sets
in the architype and recovered files, and verify that the metadata
in the versions in the recovered file (if they appear at all), agree
with the architype versions where it must, and contain plausible
values where it is possible that changes were lost.
The test also looks at the raw data, and dumps the architype and
recovered versions to stdout if anything looks "odd".
At present, my addition only works with the integer chunked
data set -- not with the other data set types that can be
created by trecover.
The code for the verify function is in the new file
trecover_verifier.c. Updated Makefile.am and the manifest
accordingly.
5) Modified the synchronous crash test in tools/h5recover to
to function when return codes are not passed back to the
calling script, and to use the above modifications to trecover
to examine the recovered file, instead of comparing the output
of dumps of the architype and recovered files.
6) Commented out the asynchronous crash test in tools/h5recover,
as the functionality of that test is now handled in the
"walking crash" test.
7) Modified the "walking crash" test to use the trecover modifications
to verify each recovery.
8) Modified the journaling file marking tests to function when the
the return code is not passed back to the calling script.
9) Commented out the "tgroup-1.ls 1 -w80 -r -g tgroup.h5" in
tools/h5ls. I am given to understand that this test was
failing on redstorm due to yod's failure to pass back return
codes. I have not investigated this personally.
10) Updated bin/reconfigure to deal with recent changes in the file
system structure on jam.
Testing:
Tested (serial) on Phoenix, Linew, and RSQ -- all pass. Note that
on the "walking crash" test in tools/h5recover, I was unable to
set the asynchronous crash delay small enough to get the crash to
occur before trecover completed (I got down to 1 usec). This
was not a problem on on redstorm the last time we tried testing
there, so I'm not too worried about it.
I also did a parallel test on jam -- this test failed with an assertion
failure in dtypes -- output follows:
============================
dtypes Test Log
============================
Testing non-aligned conversions (ALIGNMENT=1)....
Testing H5Tget_class() PASSED
Testing H5Tcopy() PASSED
Testing H5Tdetect_class() PASSED
Testing compound datatypes PASSED
Testing query functions of compound and enumeration types PASSED
Testing transient datatypes PASSED
Testing named datatypes PASSED
Testing functions of encoding and decoding datatypes PASSED
Testing encoding datatypes with the 'use the latest format' flag PASSED
Testing exceptions for int <-> float conversions PASSED
Testing deprected API routines for datatypes PASSED
Testing string conversions PASSED
Testing random string conversion speed PASSED
Testing some type functions for string PASSED
Testing compound element reordering PASSED
Testing compound subset conversions PASSED
Testing compound element shrinking & reordering PASSED
Testing optimized struct converter PASSED
Testing compound element growing PASSED
Testing compound element insertion PASSED
Testing packing compound datatypes PASSED
Testing compound datatype with VL string dtypes: H5FD.c:2150: H5FD_write: Assertion `1==H5P_isa_class(dxpl_id,(H5P_CLS_DATASET_XFER_g))' failed.
Command terminated by signal 6
0.20user 0.10system 0:00.39elapsed 75%CPU (0avgtext+0avgdata 0maxresident)k
0inputs+0outputs (4major+10263minor)pagefaults 0swaps
I'm checking in anyway, as this looks unrelated to any of my recent
changes. Quincey and I should get together about this one.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Modified H5C2_journal_post_flush() write the super block and flush
the file before truncating the journal. Failure to do this opened
a window in which the application could crash leaving the HDF5 file
in a state that was un-recoverable.
The hope is that this will fix the file recovery bug observed on
RSQ -- but I have not been able to test there. However, I was able
to generate a similar bug on Linew, and this fix seems to deal with
the Linew bug.
Added a third test to the h5recovery tests. This is really
a test for the library, but it was easier to use existing test
code there to construct the new test.
The new test runs the same application repeatedly, but setting a
timer to crash the application at progressively later times. The object
is to search for windows in which the application leaves the HDF5 file
in an un-recoverable state.
Also, updated H5recover.c to use HDstrtoll() instead of HDstrtod()
to read some addresses and such from the journal file.
Tested serial (debug) on Phoenix and Linew, and parallel (debug)
on Jam.
|
|
|
|
| |
dependent.
|
|
|
|
|
|
|
|
| |
for the
async test.
Tested: kagiso.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
cache2_journal.c
There have been quite a few checkins while I have been testing this
fix and other issues, but I'm going ahead with this checking anyway,
as my changes are quite localized.
While testing my fix, I ran into two other problems that seem to
have appeared since my last checkin:
First, in a parallel build on Kagiso, we have started getting an
assertion failure in the dtypes test -- error message follows below:
--------------------------------------------------
============================
Testing dtypes
============================
dtypes Test Log
============================
Testing non-aligned conversions (ALIGNMENT=1)....
Testing H5Tget_class() PASSED
Testing H5Tcopy() PASSED
Testing H5Tdetect_class() PASSED
Testing compound datatypes PASSED
Testing query functions of compound and enumeration types PASSED
Testing transient datatypes PASSED
Testing named datatypes PASSED
Testing functions of encoding and decoding datatypes PASSED
Testing encoding datatypes with the 'use the latest format' flag PASSED
Testing exceptions for int <-> float conversions PASSED
Testing deprected API routines for datatypes PASSED
Testing string conversions PASSED
Testing random string conversion speed PASSED
Testing some type functions for string PASSED
Testing compound element reordering PASSED
Testing compound subset conversions PASSED
Testing compound element shrinking & reordering PASSED
Testing optimized struct converter PASSED
Testing compound element growing PASSED
Testing compound element insertion PASSED
Testing packing compound datatypes PASSED
Testing compound datatype with VL string dtypes: H5F
D.c:2150: H5FD_write: Assertion `1==H5P_isa_class(dxpl_id,(H5P_CLS_DATASET_XFER_g
))' failed.
Command terminated by signal 6
-------------------------------------------------------
Second, while Mike M's recent checkin of a fix to h5recover took care of the
failure that Albert was seeing, it exposed another problem with h5recover.
Specifically, Mike is using HDstrtod to parse addresses and eoa's. If I
read the man page on strtod correctly, this should work in C99, but it
appears that for whatever reason, we are not getting the C99 version of
this library call in the Linew build.
A possible short term fix would be to change the Solaris configuration to
use C99 (if that is possible), but I'm not sure we want to do this, as
strtod() returns a double, which will not in general have sufficient
precision to store an haddr_t in its mantissa.
Alternatively, we could add strtoull() to our list of C library routines
that we are allowed to use (i.e. implement HDstrtoull()).
My personal inclination is to write our own address parsing routine, and
not mess with the different versions of C.
I'm bouncing this decision up to Albert and/or Quincey, and thus I'm
leaving this problem alone for now.
Tested on Phoenix (serial)
Kagiso (serial and parallel)
Linew (serial)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Bring revisions 15289:15457 from trunk into metadata journaling
branch.
Tested on:
FreeBSD/32 6.2 (duty) in debug mode
FreeBSD/64 6.2 (liberty) w/C++ & FORTRAN, in debug mode
Linux/32 2.6 (kagiso) w/PGI compilers, w/C++ & FORTRAN, w/threadsafe,
in debug mode
Linux/64-amd64 2.6 (smirom) w/default API=1.6.x, w/C++ & FORTRAN,
in production mode
Linux/64-ia64 2.6 (cobalt) w/Intel compilers, w/C++ & FORTRAN,
in production mode
Solaris/32 2.10 (linew) w/deprecated symbols disabled, w/C++ & FORTRAN,
w/szip filter, in production mode
Mac OS X/32 10.5.2 (amazon) in debug mode
Linux/64-ia64 2.4 (tg-login3) w/parallel, w/FORTRAN, in production mode
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Review and get code to conform to standard library coding style, also
add 'const' keyword to "set" routines.
Tested on:
FreeBSD/32 6.2 (duty) in debug mode
FreeBSD/64 6.2 (liberty) w/C++ & FORTRAN, in debug mode
Linux/32 2.6 (kagiso) w/PGI compilers, w/C++ & FORTRAN, w/threadsafe,
in debug mode
Linux/64-amd64 2.6 (smirom) w/default API=1.6.x, w/C++ & FORTRAN,
in production mode
Linux/64-ia64 2.6 (cobalt) w/Intel compilers, w/C++ & FORTRAN,
in production mode
Solaris/32 2.10 (linew) w/deprecated symbols disabled, w/C++ & FORTRAN,
w/szip filter, in production mode
Mac OS X/32 10.5.2 (amazon) in debug mode
Linux/64-ia64 2.4 (tg-login3) w/parallel, w/FORTRAN, in production mode
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Remove remnant of initial H5Pset_journal() routine
Tested on:
FreeBSD/32 6.2 (duty) in debug mode
FreeBSD/64 6.2 (liberty) w/C++ & FORTRAN, in debug mode
Linux/32 2.6 (kagiso) w/PGI compilers, w/C++ & FORTRAN, w/threadsafe,
in debug mode
Linux/64-amd64 2.6 (smirom) w/default API=1.6.x, w/C++ & FORTRAN,
in production mode
Linux/64-ia64 2.6 (cobalt) w/Intel compilers, w/C++ & FORTRAN,
in production mode
Solaris/32 2.10 (linew) w/deprecated symbols disabled, w/C++ & FORTRAN,
w/szip filter, in production mode
Mac OS X/32 10.5.2 (amazon) in debug mode
Linux/64-ia64 2.4 (tg-login3) w/parallel, w/FORTRAN, in production mode
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Add return value to metadata journaling status change callback routines,
so we can detect errors in them.
Tested on:
FreeBSD/32 6.2 (duty) in debug mode
FreeBSD/64 6.2 (liberty) w/C++ & FORTRAN, in debug mode
Linux/32 2.6 (kagiso) w/PGI compilers, w/C++ & FORTRAN, w/threadsafe,
in debug mode
Linux/64-amd64 2.6 (smirom) w/default API=1.6.x, w/C++ & FORTRAN,
in production mode
Linux/64-ia64 2.6 (cobalt) w/Intel compilers, w/C++ & FORTRAN,
in production mode
Solaris/32 2.10 (linew) w/deprecated symbols disabled, w/C++ & FORTRAN,
w/szip filter, in production mode
Mac OS X/32 10.5.2 (amazon) in debug mode
Linux/64-ia64 2.4 (tg-login3) w/parallel, w/FORTRAN, in production mode
|
|
|
|
|
|
|
|
| |
Description: Correcting a typo which was causing the wrong value to get
pulled in for journal entry size, occasionally causing entries
to become truncated and only partially journaled.
Tested: kagiso
|
|
|
|
|
|
|
|
|
| |
expected output.
But the async test still fails as h5recover fails to recover the file to
a legal state that is acceptable by h5dump.
The rest of the other tests passed.
|
|
|
|
|
|
|
|
|
|
| |
Description: The pre-parse of the journal file to determine maximum entry
size (to determine proper memory allocation amount) didn't take
into account incomplete entries at the end of the file. Modified
to only parse up through the last complete entry, rather than
the entire file.
Tested: kagiso
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Use metadata journaling callback to allow dataset code to track journal
status changes and flush cached info appropriately.
Tested on:
FreeBSD/32 6.2 (duty) in debug mode
FreeBSD/64 6.2 (liberty) w/C++ & FORTRAN, in debug mode
Linux/32 2.6 (kagiso) w/PGI compilers, w/C++ & FORTRAN, w/threadsafe,
in debug mode
Linux/64-amd64 2.6 (smirom) w/default API=1.6.x, w/C++ & FORTRAN,
in production mode
Linux/64-ia64 2.6 (cobalt) w/Intel compilers, w/C++ & FORTRAN,
in production mode
Solaris/32 2.10 (linew) w/deprecated symbols disabled, w/C++ & FORTRAN,
w/szip filter, in production mode
Mac OS X/32 10.5.2 (amazon) in debug mode
Linux/64-ia64 2.4 (tg-login3) w/parallel, w/FORTRAN, in production mode
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Convert local heap cache client to use metadata journaling cache.
Other minor cleanups & simplifications, etc.
Tested on:
FreeBSD/32 6.2 (duty) in debug mode
FreeBSD/64 6.2 (liberty) w/C++ & FORTRAN, in debug mode
Linux/32 2.6 (kagiso) w/PGI compilers, w/C++ & FORTRAN, w/threadsafe,
in debug mode
Linux/64-amd64 2.6 (smirom) w/default API=1.6.x, w/C++ & FORTRAN,
in production mode
Linux/64-ia64 2.6 (cobalt) w/Intel compilers, w/C++ & FORTRAN,
in production mode
Solaris/32 2.10 (linew) w/deprecated symbols disabled, w/C++ & FORTRAN,
w/szip filter, in production mode
Mac OS X/32 10.5.2 (amazon) in debug mode
Linux/64-ia64 2.4 (tg-login3) w/parallel, w/FORTRAN, in production mode
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
journaling in progress to avoid calls to fork(). Did this by creating
cache2_jnl_file_marking (a program that sets up or checks the results
of the specified file marking test depending on parameters passed to
it) and test/testjnlfilemarking.sh (a shell script to call
cache2_jnl_file_marking and report results).
Also fixed an input validation bug in src/H5AC.c in passing.
Tested on Phoenix (serial -- debug and production mode)
Kagiso (parallel)
Linew (serial)
There was another checkin during these tests. As the changes looked
orthoginal to mine, I updated and retested on Phoenix (serial / debug)
only before this checkin.
|
|
|
|
|
|
|
|
|
|
|
| |
Description: Converted the global heap metadata cache clients over to
use the new journaling cache callbacks.
Separated cache clients into new H5HGcache.c file. Added
H5HGcache.c to MANIFEST, added into src/Makefile.am, and
ran bin/reconfigure to regenerate Makefile.in file.
Tested: kagiso, smirom, linew, duty
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
revised cache.
Note that this conversion is not as efficient as it should be. Specifically,
it does it more memcpy's between the metadata cache's on disk image of the
direct block and the fractal heap's on disk image of the direct block than
is absolutely essential. Eventually, we will want to fix this -- probably
by allowing the metadata cache and the fractal heap direct block to share
a common on disk image of the direct block. However, this will require
extensions to the client / metadata cache interface, and some reworking of
the fractal heap as well.
This checkin also includes Mike M's fix to the Linew specific bug mentioned
in my checkin of 22 Aug 2008.
Tested on Phoenix (serial debug and production),
Kagiso (parallel), and
Linew (serial)
|
|
|
|
| |
Journaling feature.
|
| |
|
|
|
|
|
|
|
| |
Moved them from
$bin_PROGRAMS which are installed to $check_PROGRAMS which are built always
but not necessary tested automatically.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
/bin/tar: Argument list too long.
After some digging, I found that the tar arguments list generated from
MANIFEST was over 138940 bytes long. This was due to both more number
of files (2153) in the manifest to be distributed and the version name
(hdf5-1.9.8-metadata_journaling_a1) was a bit long. I changed the version
name to hdf5-1.9.8-MDJ_a1 which is 16 bytes shorter, resulting the argument
list to be 34448 bytes shorter. That seems acceptable to kagiso as the
tar command ran successfully.
So, I abbreviated the version name. This is a temporary fix. A real fix
would be to use consecutive tar commands, each with a reasonably shorter
list of arguments to generate the final tar file.
|
| |
|
|
|
|
|
|
| |
counter value
and continue adding more rows from there.
|
|
|
|
|
|
| |
prepare for
an Alpha1 release.
|
|
|
|
|
|
|
|
|
|
|
|
| |
a separated function for easier identification.
Removed the redundent call of H5Pset_libver_bounds().
Changed return(-1) to return(1) since main can't return negative
value, at least not in Unix systems.
-This line, and those below, will be ignored--
M enable_journaling.c
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
1) Removed references to H5AC2__CURR_JNL_CONFIG_VER from H5Fget_jnl_config()
and H5Pget_jnl_config(), and also references to
H5AC__CURR_CACHE_CONFIG_VERSION from H5Fget_mdc_config() and
H5Pget_mdc_config().
2) Removed H5Pset_journal() from H5C2journal.c, and modified test
code to use H5F/Pget/set_mdj_config() instead.
3) Implemented support for callbacks on metadata journaling status change
along with the associated registration / deregistration calls and
associated test code.
4) Fixed bug in journaling shutdown exposed by 3 above.
Tested and passed on Phoenix (serial), Linew (serial), and Kagiso (parallel).
However, while I was testing there were a couple of checkins, forcing
an update and second round of testing.
On the second round, tested and passed on Phoenix (serial) and
Kagiso (parallel), but failed on Linew (serial).
As best I can tell, this was caused by Mike M's checkin -- which
broke the smoke checks in cache2_journal on Linew but not Phoenix
or Kagiso. A typical delta in the architype files follows:
linew.hdfgroup.uiuc.edu% diff -ctw cache2_journal_sc00_000.jnl tmp/cache2_journal_sc00_000.jnl
*** cache2_journal_sc00_000.jnl Fri Aug 22 08:28:49 2008
--- tmp/cache2_journal_sc00_000.jnl Fri Aug 22 05:08:41 2008
***************
*** 1,5 ****
! 0 ver_num 1 target_file_name cache_journal_test.h5 creation_date Fri Aug 22 human_readable 1
! E eoa_value 0x0
C comment Begin transaction on transaction 1.0.
1 bgn_trans 1
2 trans_num 1 length 1 base_addr 0x401 body 01
--- 1,5 ----
! 0 ver_num 1 target_file_name cache_journal_test.h5 creation_date Wed Aug 20 human_readable 1
! E eoa_value 0x772a9c01
C comment Begin transaction on transaction 1.0.
1 bgn_trans 1
2 trans_num 1 length 1 base_addr 0x401 body 01
As you can see, it looks like garbage is getting into the first
eoa write on Linew.
I'm checking in anyway, as Quincey needs my changes, and I will not
have time to work on this for several days.
Mike: Let me know if you are tackling this one -- if not, I'll deal with it.
|
|
|
|
|
|
|
|
| |
flush to
allow manual abort of the application, thus simulate a crash. If program
is NOT aborted, it will complete successfully. Previously, it would
abort itself for -w but that is removed.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
- EOA logging update
Description:
- EOA values will now be written to the journal
file in their own transaction when the EOA
changes.
- The EOA will be udpated in the HDF5 file's
superblock before the recovery process begins.
This should prevent some loss of raw data as the
file won't be getting truncated upon file open as
it will read the correct EOA value from the
superblock.
- Removed storing of EOA in journal entry messages
since they're in their own transaction.
- Updated tests to reflect change of transaction
formats. Regenerated smoke test files to account
for new entry types, and tweaked transaction number
tests to reflect change in size of journal entries.
- Large testfiles (in test/testfiles) should now
unzip when ./configure is run.
- When journal file is supplied but contains no
complete transactions, instead of reporting
an error, h5recover now informs the user of said
nonexistant transactions, and opens/closes the
hdf5 file with the journal recovered flag set.
- Other various organizational changes to h5recover,
included a bit more added to verbose output.
Tested:
- kagiso, smirom
|
|
|
|
| |
Renamed the reopen mode (-r) to write mode (-w) which is more intutive.
|
|
|
|
|
|
| |
parameters.
They does not affect the results and are not needed.
|
|
|
|
|
|
|
| |
Description: converted the fractal heap header and indirect block metadata
cache clients over to use the new journaling cache callbacks.
Tested: kagiso, smirom
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
from the H5AC2_cache_config_t structure and the associated
H5P/H5Fget/set_mdc_config() API calls, and into the newly created
H5AC2_jnl_config_t structure and H5P/H5Fget/set_jnl_config() calls.
Updated test code accordingly.
Updated the trace file test code for journaling.
Also folded in a fix to an assertion bug in H5C2pkh.h
Tested serial on Phoenix and Linew, and parallel (with and without the
trace file enabled) on kagiso.
|
|
|
|
|
|
|
|
|
|
| |
addresses
for easier octo dump of data file.
Moved latest lib version to apply all the time.
Tested:
kagiso.
|
|
|
|
|
|
| |
Added -z option to turn OFF all caching.
Tested: kagiso.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
- Bug fixing and cleanup
Description:
- Fixed the problem where the recovery wasn't "sticking" on the first pass
of using the tool. Now should only need to run the tool once to recover
a file.
- Uncommented out some lines generating more verbose output, so now
entry size and addresses are visible with -v (--verbose) option.
- Fixed off-by-one memory allocation error.
- Cleaned up comments to allow for easier reading.
Tested:
- kagiso, smirom
|
|
|
|
|
|
|
|
| |
Description: converted the shared object header message and index stored
as a list metadata cache clients over to use the new journaling
cache callbacks.
Tested: kagiso, smirom
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Convert object header cache client to use the new metadata journaling
cache, which included adding a new client for handling continuation chunks.
Added "real" protect calls around modifying chunks in object headers.
Switched a few more metadata cache library API routines to drop the
file pointer, when it is not needed (pinning/unpinning entries, etc.)
Fixed bug in journaling cache handling of 'image_len' callbacks and
also changed cache to retry deserializing entries when the entry's size is
larger than the speculative size initially tried.
Retrying for 'image_len' callbacks has problems with the 'multi' VFD,
so the h5dump and FORTRAN 'multi' tests are commented out, until the changes to
the 'multi' VFD from the file free space branch are brought back into the
trunk.
Currently, the 'h5recover' tool has a bug which requires it to be run
twice before replaying the journal "sticks". However, this is from an earlier
checkin, since the code in the branch already has this behavior... :-(
Tested on:
FreeBSD/32 6.2 (duty) in debug mode
FreeBSD/64 6.2 (liberty) w/C++ & FORTRAN, in debug mode
Linux/32 2.6 (kagiso) w/PGI compilers, w/C++ & FORTRAN, w/threadsafe,
in debug mode
Linux/64-amd64 2.6 (smirom) w/default API=1.6.x, w/C++ & FORTRAN,
in production mode
Linux/64-ia64 2.6 (cobalt) w/Intel compilers, w/C++ & FORTRAN,
in production mode
Solaris/32 2.10 (linew) w/deprecated symbols disabled, w/C++ & FORTRAN,
w/szip filter, in production mode
Mac OS X/32 10.5.4 (amazon) in debug mode
Linux/64-ia64 2.4 (tg-login3) w/parallel, w/FORTRAN, in production mode
|
|
|
|
|
|
|
|
|
| |
journaling,
along with test code that uncovered the problem and associated additions &
changes to the gziped test files in test/testfiles
Tested serial on Phoenix and parallel on kagiso
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
which
I broke in the last checkin.
Also added test to verify that the journal_recovered flag in fact
causes a file that is marked as having journaling in progress will
cause the file to be unmarked.
Tested serial on Phoenix and parallel on Kagiso. h5recover is still
failing, but everything else seems to pass.
|
|
|
|
|
|
|
|
| |
identify
this is a feature branch.
H5committested.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
specifically:
1) Fix for failure to detect journaling in progress on HDF5 files which
were not closed correctly. Also associated test code.
Note that this required addition of code to test for journaling in
progress and enable journaling at the end of H5F_flush(). In passing,
I was able to get rid of the wacky code that queued journaling setup
at cache creation time.
2) Test code for startup and shutdown of journaling on an open file.
3) Updates to start checking journal output against architype files
instead of just generating architypes at test time.
Note that per Quincey's request, I have checked in gziped versions
of the architype files. At some point, we will have to add code
to automatically unzip these files, but for the time being you will
have to go to test/testfiles and "gunzip *.gz". The journal tests
will still pass if you don't, but you will get a warning about
missing test files.
4) Fixed bug in journal entry logging code that allowed a comment to
appear in the journal file before the journal file header. (Mike M,:
Please review my fix to verify that I haven't clobbered anything.)
5) Additional test code.
Note that more test code would be a good idea, but this set of bug
fixes should be enough to get us through the basic demo -- at least
as far as the metadata cache is concerned.
Tested serial on Phoenix, and parallel on Kagiso. Also, tested serial
on Linew just prior to some last minute minor edits.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Disable dataset caching when journaling is enabled.
Tested on:
FreeBSD/32 6.2 (duty) in debug mode
FreeBSD/64 6.2 (liberty) w/C++ & FORTRAN, in debug mode
Linux/32 2.6 (kagiso) w/PGI compilers, w/C++ & FORTRAN, w/threadsafe,
in debug mode
Linux/64-amd64 2.6 (smirom) w/default API=1.6.x, w/C++ & FORTRAN,
in production mode
Linux/64-ia64 2.6 (cobalt) w/Intel compilers, w/C++ & FORTRAN,
in production mode
Solaris/32 2.10 (linew) w/deprecated symbols disabled, w/C++ & FORTRAN,
w/szip filter, in production mode
Mac OS X/32 10.5.2 (amazon) in debug mode
Linux/64-ia64 2.4 (tg-login3) w/parallel, w/FORTRAN, in production mode
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Bring revisions 15210:15289 from the trunk into the metadata journaling branch.
Tested on:
FreeBSD/32 6.2 (duty) in debug mode
FreeBSD/64 6.2 (liberty) w/C++ & FORTRAN, in debug mode
Linux/32 2.6 (kagiso) w/PGI compilers, w/C++ & FORTRAN, w/threadsafe,
in debug mode
Linux/64-amd64 2.6 (smirom) w/default API=1.6.x, w/C++ & FORTRAN,
in production mode
Linux/64-ia64 2.6 (cobalt) w/Intel compilers, w/C++ & FORTRAN,
in production mode
Solaris/32 2.10 (linew) w/deprecated symbols disabled, w/C++ & FORTRAN,
w/szip filter, in production mode
Linux/64-ia64 2.4 (tg-login3) w/parallel, w/FORTRAN, in production mode
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Merge revisions 15130:15210 from trunk into metadata journaling branch
Tested on:
FreeBSD/32 6.2 (duty) in debug mode
FreeBSD/64 6.2 (liberty) w/C++ & FORTRAN, in debug mode
Linux/32 2.6 (kagiso) w/PGI compilers, w/C++ & FORTRAN, w/threadsafe,
in debug mode
Linux/64-amd64 2.6 (smirom) w/default API=1.6.x, w/C++ & FORTRAN,
in production mode
Linux/64-ia64 2.6 (cobalt) w/Intel compilers, w/C++ & FORTRAN,
in production mode
Solaris/32 2.10 (linew) w/deprecated symbols disabled, w/C++ & FORTRAN,
w/szip filter, in production mode
Linux/64-ia64 2.4 (tg-login3) w/parallel, w/FORTRAN, in production mode
|
|
|
|
|
|
|
|
| |
Description: converted the v2 b-tree header, internal node, and leaf node
metadata cache clients over to use the new journaling cache
callbacks.
Tested: kagiso, smirom
|
|
|
|
|
|
| |
Added -p for patch mode that is needed temporary.
Tested: kagiso
|
|
|
|
|
|
| |
view the h5dump output.
Tested: kagiso.
|
|
|
|
|
|
|
|
|
|
| |
because the journal
file is already existing.
Remove the journal file before starting the journaling tests.
Tested: h5cmmittested.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Merge revisions 15037:15130 from trunk into metadata journaling branch
Tested on:
FreeBSD/32 6.2 (duty) in debug mode
FreeBSD/64 6.2 (liberty) w/C++ & FORTRAN, in debug mode
Linux/32 2.6 (kagiso) w/PGI compilers, w/C++ & FORTRAN, w/threadsafe,
in debug mode
Linux/64-amd64 2.6 (smirom) w/default API=1.6.x, w/C++ & FORTRAN,
in production mode
Linux/64-ia64 2.6 (cobalt) w/Intel compilers, w/C++ & FORTRAN,
in production mode
Solaris/32 2.10 (linew) w/deprecated symbols disabled, w/C++ & FORTRAN,
w/szip filter, in production mode
Mac OS X/32 10.5.2 (amazon) in debug mode
Linux/64-ia64 2.4 (tg-login3) w/parallel, w/FORTRAN, in production mode
|