| Commit message (Collapse) | Author | Age | Files | Lines |
|
|
|
|
|
| |
2) Skip dirty corked entries when doing auto ageout entries in the cache.
Tested on koala, jam, platypus, ostrich.
PENDING: code review and more cache related tests.
|
|
|
|
| |
ball for the customer.
|
|
|
|
|
|
| |
1.9.170-swmr_chksum3 tar ball.
Platforms tested: jam
|
|
|
|
|
|
|
|
|
| |
routines.
Tested on jam, koala, ostrich, platypus.
PENDING:
(1) Code review.
(2) More work on tests, comments, cache related work.
|
|
|
|
|
|
| |
(A) SWMR related public routines: H5Fstart_swmr_write, H5Pget/set_append_flush, H5Pget/set_object_flush_cb.
(B) File locking.
Tested on jam, koala, ostrich, platypus.
|
|
|
|
| |
h5committested.
|
|
|
|
| |
SWMR Phase II task 1
|
|
|
|
|
|
|
|
|
| |
Review Vailin's "metadata retry" code and make various cleanups to it.
Also, disallow SWMR access for non-latest format and for non-POSIX VFDs.
Tested on:
Mac OSX/64 10.9.0 (amazon) w/gcc 4.8 and make check-vfd
|
| |
|
|
|
|
| |
ball for DLS.
|
|
|
|
| |
h5committested.
|
|
|
|
| |
h5committested.
|
| |
|
| |
|
|
|
|
|
|
|
| |
checksum.
Implement new public routine H5Fget_metadata_read_retries_info().
h5committested.
|
|
|
|
|
|
|
|
| |
this
branch can be identified with this unique string.
Tested: Koala passed.
|
|
|
|
| |
h5committested.
|
|
|
|
|
|
|
| |
Changes the MD cache so that writing v-1 B-tree nodes under SWMR
semantics will fail.
Tested on: jam (previously tested on more platforms)
|
|
|
|
|
|
|
|
|
|
|
| |
is opened with SWMR access.
There are debugging printfs which will be removed when coding is finalized.
Also some bug fixes:
1) accum.c--clean up some warning messages and use new_argv/new_envp for the call to execve.
2) hl/tools/h5watch--clean up some warning messages and a bug fix for h5watch.c.
This checkin is awaiting code review feedback.
|
|
|
|
| |
h5committested.
|
|
|
|
| |
h5committested.
|
|
|
|
| |
h5committested.
|
|
|
|
| |
h5committested.
|
|
|
|
| |
h5committested.
|
|
|
|
| |
h5committested.
|
|
|
|
| |
h5committested.
|
|
|
|
| |
h5committested.
|
|
|
|
| |
h5committested.
|
|
|
|
| |
h5committested.
|
|
|
|
| |
h5committested.
|
|
|
|
| |
h5committested.
|
|
|
|
| |
h5committested.
|
|
|
|
|
|
| |
1) H5O_load() in H5Ocache.c: when reading a block that is > spec_read siez, read the whole block in again and possibly decode the header.
2) H5F_accum_write() in H5Faccum.c: for a large write that is >= H5F_ACCUM_MAX_SIZE, flush the metadata in the accumulator first before the write.
Tests are added to test/ohdr.c and test/accum.c. h5committested.
|
|
|
|
| |
h5committtested.
|
|
|
|
| |
h5committested.
|
|
|
|
| |
h5committested.
|
|
|
|
| |
h5committested.
|
|
|
|
| |
h5committested.
|
|
|
|
| |
h5committested.
|
|
|
|
| |
h5committested.
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Descriptions:
When removing object header messages, it is possible for object header
continuation messages to move to a different chunk. When this happens, flush
dependencies need to be updated to reflect the new structure. This change adds
code to update the flush dependencies, and a test for this.
Also fixed a bug where the flush dependency no the object header proxy was not
being destroyed when an object header chunk was deleted.
Tested: ummon
|
|
|
|
| |
Tested: ummon
|
|
|
|
|
|
|
|
|
|
|
| |
Description:
Added flush dependencies between object header chunks. Also added a new object
header proxy object, which is a child dependency of all object header chunks.
All objects which are child dependencies of an object header should set up a
flush dependency on the object header proxy, so the dependency applies to all
object header chunks.
Tested: ummon; jam (fails ph5diff -v, unrelated), koala, ostrich (h5committest)
|
|
|
|
| |
Tested: ummon
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
multiple parents
Description:
Reworked how flush dependencies worked internally, allowing multiple flush
dependency parents, and removing the notion of flush dependency heights, instead
keeping track of the number of dirty descendents to determine if parents can be
flushed. Also removed the requirement that cache clients destroy flush
dependencies before eviction (this is now handled by the cache) and removed the
maximum number of passes (the cache should detect infinite loops elsewhere).
Added test cases for this.
Tested: durandal
|
|
|
|
|
|
|
| |
Add missing skip list header to H5AC.c
Tested on:
jam (trivial change)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Description:
Since the "flush me last" code was introduced, it was no longer possible for the
superblock to be flushed to disk without flushing the entire file. This made it
impossible to open a file concurrently, even after calling H5Gflush on the root
group. The superblock was originally supposed to be flushed every time any
object was explicitly flushed with H5*flush.
Modified H5C_flush_cache to ignore flush me last flag if the entry is marked for
flush (which is done either by a tag for H5*flush or to free up space in cache)
and the flush marked entries flag is set.
Tested: ummon
|
|
|
|
|
|
|
| |
Updated CMakelists.txt to include H5FSint.c
Tested on:
jam
|
|
|
|
|
|
|
| |
Updated CMakelists.txt to include H5FSint.c
Tested on:
jam
|
|
|
|
|
|
|
|
| |
Converted an H5E_THROW to HGOTO_ERROR in the free space manager
code.
Tested on:
jam
|