greg/stasis-bLSM
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions
147 commits 1 branch 0 tags 670 KiB
Commit graph

53 commits

Author SHA1 Message Date
sears
a1f3f92cca added bulk-insert APIs, prevent c1-c2 from outrunning c0-c1 by too much, and fix the overshoot computations in merge manager (roughly corresponds to run 20) 
git-svn-id: svn+ssh://svn.corp.yahoo.com/yahoo/yrl/labs/pnuts/code/logstore@1074 8dad8b1f-cf64-0410-95b6-bcf113ffbcfe
 2010-08-30 22:22:25 +00:00
sears
960ff041e2 remove old comments 
git-svn-id: svn+ssh://svn.corp.yahoo.com/yahoo/yrl/labs/pnuts/code/logstore@1055 8dad8b1f-cf64-0410-95b6-bcf113ffbcfe
 2010-08-24 00:40:48 +00:00
sears
07b35f4e20 scattered, and poorly tested fixes. finer grained latching for tree nodes; throttle c0-c1 merge when c0 is small; fix delta computations for c0 throttling; disable short pauses for c1-c2 merge. this gets the bulk load throughput up to 8500, but workloada is still slow 
git-svn-id: svn+ssh://svn.corp.yahoo.com/yahoo/yrl/labs/pnuts/code/logstore@1045 8dad8b1f-cf64-0410-95b6-bcf113ffbcfe
 2010-08-21 03:09:18 +00:00
sears
b894cebaf7 add memTreeComponent::batchedRevalidatingIterator; amortize cost of red black latching during merge 
git-svn-id: svn+ssh://svn.corp.yahoo.com/yahoo/yrl/labs/pnuts/code/logstore@1017 8dad8b1f-cf64-0410-95b6-bcf113ffbcfe
 2010-08-18 17:29:25 +00:00
sears
1e487bbc54 re-enable old merge policy for c1-c2 merger; implement shutdown (sort of) 
git-svn-id: svn+ssh://svn.corp.yahoo.com/yahoo/yrl/labs/pnuts/code/logstore@1006 8dad8b1f-cf64-0410-95b6-bcf113ffbcfe
 2010-08-17 21:23:39 +00:00
sears
dc75f6d1f0 Implemented snowshoveling, with caveats. 
This commit includes concurrency control for datapages, the new backpressure for the c0-c1 merger, and (untested) support for defining NO_SNOWSHOVEL, and using the old approach.

There are a number of important limitations:
 - statistics gathering is somewhat messed up
 - the c0-c1 merge thread polls for new data
 - on a sequential write, if the writers stall, the merge thread could empty memory and initiate an unnecessary merge.  Instead, we should never finish a merge if c0 is less than N% full
 - this commit introduces yet another batch of nasty mutex contention issues



git-svn-id: svn+ssh://svn.corp.yahoo.com/yahoo/yrl/labs/pnuts/code/logstore@955 8dad8b1f-cf64-0410-95b6-bcf113ffbcfe
 2010-08-05 17:43:46 +00:00
sears
cf6c6e5955 incease UPDATE_PROGRESS_DELTA by 10x, and FORCE_INTERVAL by 2.5x; tick no longer grabs a writelock on header_mut, and instead uses its own mutex 
git-svn-id: svn+ssh://svn.corp.yahoo.com/yahoo/yrl/labs/pnuts/code/logstore@870 8dad8b1f-cf64-0410-95b6-bcf113ffbcfe
 2010-06-21 22:59:05 +00:00
sears
fded91fec6 two unrelated changes: release rb_mut before getting header_mut on read path (breaks delta tuples), and implement first-cut support for pulling in statistics when reopening logstores (required for correct recovery) 
git-svn-id: svn+ssh://svn.corp.yahoo.com/yahoo/yrl/labs/pnuts/code/logstore@868 8dad8b1f-cf64-0410-95b6-bcf113ffbcfe
 2010-06-21 20:03:05 +00:00
sears
2530009ec0 fix deadlock; do not clear need_tick bit until overshoot is less than zero; tweak constants accordingly (fewer mutex acquistions! less blockage!) 
git-svn-id: svn+ssh://svn.corp.yahoo.com/yahoo/yrl/labs/pnuts/code/logstore@852 8dad8b1f-cf64-0410-95b6-bcf113ffbcfe
 2010-06-18 00:06:46 +00:00
sears
0fb644640a improved concurrency for merger. not happy with worst case write latency on my dev machine anymore (seems to be less than 2.5 seconds, up from less than 1). 
git-svn-id: svn+ssh://svn.corp.yahoo.com/yahoo/yrl/labs/pnuts/code/logstore@842 8dad8b1f-cf64-0410-95b6-bcf113ffbcfe
 2010-06-17 04:49:19 +00:00
sears
0ecd7491c7 fix minor stats bugs; pull most stats computation into update_progress. The idea is that update_progress and tick, which are easily amortized, are the only stats things that need to grab latches 
git-svn-id: svn+ssh://svn.corp.yahoo.com/yahoo/yrl/labs/pnuts/code/logstore@827 8dad8b1f-cf64-0410-95b6-bcf113ffbcfe
 2010-06-05 00:41:52 +00:00
sears
fe7ce3a785 tweak some constants to silence latency sla warnings 
git-svn-id: svn+ssh://svn.corp.yahoo.com/yahoo/yrl/labs/pnuts/code/logstore@825 8dad8b1f-cf64-0410-95b6-bcf113ffbcfe
 2010-06-03 00:12:31 +00:00
sears
eda514ddaa amortize the computation of statistics (but not mutex acuqisitions) across many tuple insertions. Also, refactor stats stuff with latch contention in mind (but it still grabs too many latches...). This commit causes check_tcpclient to emit warnings on my machine. Not sure if it will happen on production hardware 
git-svn-id: svn+ssh://svn.corp.yahoo.com/yahoo/yrl/labs/pnuts/code/logstore@824 8dad8b1f-cf64-0410-95b6-bcf113ffbcfe
 2010-06-02 21:47:58 +00:00
sears
4050475de9 more bugfixes / cleanup. latching is probably overly conservative at this point, but ycsb bulkload finally seems to work 
git-svn-id: svn+ssh://svn.corp.yahoo.com/yahoo/yrl/labs/pnuts/code/logstore@819 8dad8b1f-cf64-0410-95b6-bcf113ffbcfe
 2010-05-28 01:29:10 +00:00
sears
d981d91dfa numerous bugfixes; rate limit progress meter display 
git-svn-id: svn+ssh://svn.corp.yahoo.com/yahoo/yrl/labs/pnuts/code/logstore@816 8dad8b1f-cf64-0410-95b6-bcf113ffbcfe
 2010-05-27 23:15:24 +00:00
sears
6eecf1557a restore support for concurrent disk reads; logserver was accidentally using a findTuple function that assumed diffs. This commit uses rwlc, and requires Stasis -r1393 or better 
git-svn-id: svn+ssh://svn.corp.yahoo.com/yahoo/yrl/labs/pnuts/code/logstore@814 8dad8b1f-cf64-0410-95b6-bcf113ffbcfe
 2010-05-27 01:49:27 +00:00
sears
aa7f7189d6 admission control *almost* works; c1 sometimes blocks for a long time when a new c2 is created 
git-svn-id: svn+ssh://svn.corp.yahoo.com/yahoo/yrl/labs/pnuts/code/logstore@809 8dad8b1f-cf64-0410-95b6-bcf113ffbcfe
 2010-05-26 00:58:17 +00:00
sears
c50c14d259 admission control is now behaving properly (but there are still no concurrent writes, or bounds on how long admission control throttles a given writer) 
git-svn-id: svn+ssh://svn.corp.yahoo.com/yahoo/yrl/labs/pnuts/code/logstore@806 8dad8b1f-cf64-0410-95b6-bcf113ffbcfe
 2010-05-21 23:43:17 +00:00
sears
2a1157602a Rewrite merge thread synchronization code, update status messages, implement preliminary (and ineffective) admission control, and force write merged data every megabyte (so that prograess can be tracked by admission control. Also, refactor quite a few things. 
git-svn-id: svn+ssh://svn.corp.yahoo.com/yahoo/yrl/labs/pnuts/code/logstore@805 8dad8b1f-cf64-0410-95b6-bcf113ffbcfe
 2010-05-19 23:42:06 +00:00
sears
1adb980555 added throttling policy and better stats. the throttling makes a merge policy bug painfully obvious (committing because i plan to do some refactoring...) 
git-svn-id: svn+ssh://svn.corp.yahoo.com/yahoo/yrl/labs/pnuts/code/logstore@802 8dad8b1f-cf64-0410-95b6-bcf113ffbcfe
 2010-05-12 22:16:41 +00:00
sears
7b2560d906 initial merge manager (multi-merge) statistics 
git-svn-id: svn+ssh://svn.corp.yahoo.com/yahoo/yrl/labs/pnuts/code/logstore@797 8dad8b1f-cf64-0410-95b6-bcf113ffbcfe
 2010-04-29 23:13:04 +00:00
sears
176a1e804a more encapsulation changes 
git-svn-id: svn+ssh://svn.corp.yahoo.com/yahoo/yrl/labs/pnuts/code/logstore@794 8dad8b1f-cf64-0410-95b6-bcf113ffbcfe
 2010-04-29 00:57:48 +00:00
sears
151373d09a refactoring; move init_stasis into logtable, put merge stats into their own module 
git-svn-id: svn+ssh://svn.corp.yahoo.com/yahoo/yrl/labs/pnuts/code/logstore@792 8dad8b1f-cf64-0410-95b6-bcf113ffbcfe
 2010-04-28 19:21:25 +00:00
sears
fb073c5671 fix integer arithmetic bug that caused C0 and C1 to merge arbitrarily often 
git-svn-id: svn+ssh://svn.corp.yahoo.com/yahoo/yrl/labs/pnuts/code/logstore@767 8dad8b1f-cf64-0410-95b6-bcf113ffbcfe
 2010-03-31 22:54:26 +00:00
sears
e1c937a602 Refactoring of logtable. Cleaned a bunch of includes, and logtable is now a template. This is in preparation for the new merge policy 
git-svn-id: svn+ssh://svn.corp.yahoo.com/yahoo/yrl/labs/pnuts/code/logstore@698 8dad8b1f-cf64-0410-95b6-bcf113ffbcfe
 2010-03-17 21:51:26 +00:00
sears
28433bf336 remove logiterators.h/cpp and stop #including datapage.cpp 
git-svn-id: svn+ssh://svn.corp.yahoo.com/yahoo/yrl/labs/pnuts/code/logstore@690 8dad8b1f-cf64-0410-95b6-bcf113ffbcfe
 2010-03-13 00:41:37 +00:00
sears
c8c48a188d more refactoring; all iterators now live in their respective tree components, fixed some hardcoded parameters; encapsulate allocators inside of diskTreeComponent 
git-svn-id: svn+ssh://svn.corp.yahoo.com/yahoo/yrl/labs/pnuts/code/logstore@689 8dad8b1f-cf64-0410-95b6-bcf113ffbcfe
 2010-03-13 00:05:06 +00:00
sears
5f0d5c4f97 use RegionAllocator for InternalNodes and DataPages 
git-svn-id: svn+ssh://svn.corp.yahoo.com/yahoo/yrl/labs/pnuts/code/logstore@685 8dad8b1f-cf64-0410-95b6-bcf113ffbcfe
 2010-03-09 23:17:03 +00:00
sears
8d7199676c made diskTreeComponent more object oriented; many of its functions are now private 
git-svn-id: svn+ssh://svn.corp.yahoo.com/yahoo/yrl/labs/pnuts/code/logstore@684 8dad8b1f-cf64-0410-95b6-bcf113ffbcfe
 2010-03-09 22:18:55 +00:00
sears
54e73ab803 move the internal node region allocator stuff into diskTreeComponent::internalNode 
git-svn-id: svn+ssh://svn.corp.yahoo.com/yahoo/yrl/labs/pnuts/code/logstore@681 8dad8b1f-cf64-0410-95b6-bcf113ffbcfe
 2010-03-09 20:23:15 +00:00
sears
01c65d7a3c diskTreeComponent -> diskTreeComponent::internalNodes 
git-svn-id: svn+ssh://svn.corp.yahoo.com/yahoo/yrl/labs/pnuts/code/logstore@680 8dad8b1f-cf64-0410-95b6-bcf113ffbcfe
 2010-03-09 19:47:12 +00:00
sears
de7984b05a create memTreeComponent class. Moved memTreeIterator, rbtree_t and tearDownTree into it (still need to move changableMemTreeIterator into it...) 
git-svn-id: svn+ssh://svn.corp.yahoo.com/yahoo/yrl/labs/pnuts/code/logstore@673 8dad8b1f-cf64-0410-95b6-bcf113ffbcfe
 2010-03-09 01:42:23 +00:00
sears
632afe6d9c cleanup statistics gathering and reporting. 
constify inputs to merge functions (so they can't call freetuple())


git-svn-id: svn+ssh://svn.corp.yahoo.com/yahoo/yrl/labs/pnuts/code/logstore@662 8dad8b1f-cf64-0410-95b6-bcf113ffbcfe
 2010-03-05 19:07:47 +00:00
sears
747f4df1f3 Reopening existing trees now seems to work (no thorough tests yet). Added some network opcodes for statistics and debugging (not all are implemented yet). Reduce verbosity of server and client library. 
git-svn-id: svn+ssh://svn.corp.yahoo.com/yahoo/yrl/labs/pnuts/code/logstore@654 8dad8b1f-cf64-0410-95b6-bcf113ffbcfe
 2010-03-01 21:26:07 +00:00
sears
27720dddbe Implemented (but did not test) header update. 
git-svn-id: svn+ssh://svn.corp.yahoo.com/yahoo/yrl/labs/pnuts/code/logstore@650 8dad8b1f-cf64-0410-95b6-bcf113ffbcfe
 2010-02-27 00:35:13 +00:00
sears
7c2397340c Fixed a bunch of iterator bugs and racy merges. Added drop_database utility. 
Bugfixes:
 - Atomically deallocate regions and update the logstore object
 - Proactively invalidate iterators after each merge (before, it would simply set a not-valid bit.  This doesn't work because iterators hold page pins, which breaks force-writes)
 - Clarify semantics of opening iterators mid-stream:  All calls now return iterators that return first key >= the requested one.  revalidate() needs the key > the requested one, so it calls peek(), then (if necessary) getnext().
 - Add asserts to check that the header is latched at update, and that tuples returned by iterators are strictly monotonically increasing'
 - Improve error handling in network.h  We still get (and terminate on) SIGPIPE.

Refactoring:
 - Add dispatch function to network.h.



git-svn-id: svn+ssh://svn.corp.yahoo.com/yahoo/yrl/labs/pnuts/code/logstore@620 8dad8b1f-cf64-0410-95b6-bcf113ffbcfe
 2010-02-25 01:29:32 +00:00
sears
63a14d8509 network iterator API is now working (though it could use a better test suite) 
git-svn-id: svn+ssh://svn.corp.yahoo.com/yahoo/yrl/labs/pnuts/code/logstore@601 8dad8b1f-cf64-0410-95b6-bcf113ffbcfe
 2010-02-20 01:18:39 +00:00
sears
57c9afc8d8 rename logtree to diskTreeComponent 
git-svn-id: svn+ssh://svn.corp.yahoo.com/yahoo/yrl/labs/pnuts/code/logstore@591 8dad8b1f-cf64-0410-95b6-bcf113ffbcfe
 2010-02-19 00:59:14 +00:00
sears
2b82a62a34 broke logstore.cpp into two pieces 
git-svn-id: svn+ssh://svn.corp.yahoo.com/yahoo/yrl/labs/pnuts/code/logstore@590 8dad8b1f-cf64-0410-95b6-bcf113ffbcfe
 2010-02-19 00:52:21 +00:00
sears
ed9d4fc377 Encapsulate allocator state inside logtree. Remove redundant information from merger_args. Rewrite merge threads. This seems to fix the double free issues, and gets rid of disk space leaks. It's probably leaking ram, and does not keep the logstore headers up to date. 
git-svn-id: svn+ssh://svn.corp.yahoo.com/yahoo/yrl/labs/pnuts/code/logstore@587 8dad8b1f-cf64-0410-95b6-bcf113ffbcfe
 2010-02-18 23:31:57 +00:00
sears
ec747781ac partially remove merger_args structure 
git-svn-id: svn+ssh://svn.corp.yahoo.com/yahoo/yrl/labs/pnuts/code/logstore@579 8dad8b1f-cf64-0410-95b6-bcf113ffbcfe
 2010-02-17 23:38:31 +00:00
sears
5032e77b6c more allocator cleanups (still leaks space). Added dump_blockmap utility 
git-svn-id: svn+ssh://svn.corp.yahoo.com/yahoo/yrl/labs/pnuts/code/logstore@578 8dad8b1f-cf64-0410-95b6-bcf113ffbcfe
 2010-02-17 22:11:22 +00:00
sears
69dafd9a55 api cleanups; it is now more difficult to get confused and pass the wrong allocator into a datapage 
git-svn-id: svn+ssh://svn.corp.yahoo.com/yahoo/yrl/labs/pnuts/code/logstore@564 8dad8b1f-cf64-0410-95b6-bcf113ffbcfe
 2010-02-16 01:01:52 +00:00
sears
91273d2f43 removed alloc state from merger_args; space reclamation now seems to be working, modulo ArrayList leaks 
git-svn-id: svn+ssh://svn.corp.yahoo.com/yahoo/yrl/labs/pnuts/code/logstore@563 8dad8b1f-cf64-0410-95b6-bcf113ffbcfe
 2010-02-16 00:44:13 +00:00
sears
6f968fd825 Rewrote datapage region allocator. It no longer scribbles all over itself during merges, but it tries to double free datapage regions. Committing with datapage region reclamation disabled for now. 
git-svn-id: svn+ssh://svn.corp.yahoo.com/yahoo/yrl/labs/pnuts/code/logstore@562 8dad8b1f-cf64-0410-95b6-bcf113ffbcfe
 2010-02-15 23:02:01 +00:00
sears
d769bc0de6 fix compiler warnings. This commit makes use of an API change, and requires Stasis >= r1352 
git-svn-id: svn+ssh://svn.corp.yahoo.com/yahoo/yrl/labs/pnuts/code/logstore@561 8dad8b1f-cf64-0410-95b6-bcf113ffbcfe
 2010-02-11 21:38:05 +00:00
sears
f2df52319d Rewrote datapage. There is a feature regression: datapages do not try to extend themselves when they encounter a large record, and there is a 1:1 relationship between datapages and regions. A later revision will change the interface with the region allocator in order to address this. This will establish a 1 region -> n datapage relationship. There used to be an n:m mapping. 
git-svn-id: svn+ssh://svn.corp.yahoo.com/yahoo/yrl/labs/pnuts/code/logstore@560 8dad8b1f-cf64-0410-95b6-bcf113ffbcfe
 2010-02-11 20:04:42 +00:00
sears
940a6da6fe Reworked memory allocation and network protocol. This gargantuan commit also reduces the default size of C0, and removes quite a bit of redundant logic and error handling. 
The tests now pass, except that check_merge never terminates (it takes too long) and check_mergelarge still is not passing.

For better luck running this version of the code, turn off stasis' concurrent buffer manager.  We're doing something bad that leads to deadlocks with the concurrent buffer 
manager.  Another (the same?) bug less-frequently leads to page corruption with the old stasis buffer manager.




git-svn-id: svn+ssh://svn.corp.yahoo.com/yahoo/yrl/labs/pnuts/code/logstore@556 8dad8b1f-cf64-0410-95b6-bcf113ffbcfe
 2010-02-10 21:49:50 +00:00
sears
094a0d41ee removed some more redundant code by unifying merge_iterators 
git-svn-id: svn+ssh://svn.corp.yahoo.com/yahoo/yrl/labs/pnuts/code/logstore@532 8dad8b1f-cf64-0410-95b6-bcf113ffbcfe
 2010-01-28 02:20:49 +00:00
sears
23359c2570 fix more compiler warnings 
git-svn-id: svn+ssh://svn.corp.yahoo.com/yahoo/yrl/labs/pnuts/code/logstore@531 8dad8b1f-cf64-0410-95b6-bcf113ffbcfe
 2010-01-27 23:34:33 +00:00