* refactor leveled_sst from gen_fsm to gen_statem
* format_status/2 takes State and State Data
but this function is deprecated... put in for backward compatibility
* refactor leveled_cdb from gen_fsm to gen_statem
* disable irrelevant warning ignorer
* Remove unnecessary code paths
Only support messages, especially info messages, where they are possible.
* Mas i1820 offlinedeserialisation cbo (#403)
* Log report GC Info by manifest level
* Hibernate on range query
If Block Index Cache is not full, and we're not yielding
* Spawn to deserialise blocks offline
Hypothesis is that the growth in the heap necessary due to continual term_to_binary calls to deserialise blocks is wasting memory - so do this memory-intensive task in a short-lived process.
* Start with hibernate_after option
* Always build BIC
Testing indicates that the BIC itself is not a primary memory issue - the primary issue is due to a lack of garbage collection and a growing heap.
This change enhances the patch to offline serialisation so that:
- get_sqn & get_kv are standardised to build the BIC, and hibernate when it is built.
- the offline PId is linked to crash this process on failure (as would happen now).
* Standardise spawning for check_block/3
Now deserialise in both parts of the code.
* Only spawn for check_block if cache not full
* Update following review
* Standardise formatting
Make test more reliable. Show no new compaction after third compaction.
* Update comments
---------
Co-authored-by: Thomas Arts <thomas.arts@quviq.com>
* Mas i370 patch d (#383)
* Refactor penciller memory
In high-volume tests on large key-count clusters, so significant variation in the P0031 time has been seen:
TimeBucket PatchA
a.0ms_to_1ms 18554
b.1ms_to_2ms 51778
c.2ms_to_3ms 696
d.3ms_to_5ms 220
e.5ms_to_8ms 59
f.8ms_to_13ms 40
g.13ms_to_21ms 364
h.21ms_to_34ms 277
i.34ms_to_55ms 34
j.55ms_to_89ms 17
k.89ms_to_144ms 21
l.144ms_to_233ms 31
m.233ms_to_377ms 45
n.377ms_to_610ms 52
o.610ms_to_987ms 59
p.987ms_to_1597ms 55
q.1597ms_to_2684ms 54
r.2684ms_to_4281ms 29
s.4281ms_to_6965ms 7
t.6295ms_to_11246ms 1
It is unclear why this varies so much. The time to add to the cache appears to be minimal (but perhaps there is an issue with timing points in the code), whereas the time to add to the index is much more significant and variable. There is also variable time when the memory is rolled (although the actual activity here appears to be minimal.
The refactoring here is two-fold:
- tidy and simplify by keeping LoopState managed within handle_call, and add more helpful dialyzer specs;
- change the update to the index to be a simple extension of a list, rather than any conversion.
This alternative version of the pmem index in unit test is orders of magnitude faster to add - and is the same order of magnitude to check. Anticipation is that it may be more efficient in terms of memory changes.
* Compress SST index
Reduces the size of the leveled_sst index with two changes:
1 - Where there is a common prefix of tuple elements (e.g. Bucket) across the whole leveled_sst file - only the non-common part is indexed, and a function is used to compare.
2 - There is less "indexing" of the index i.e. only 1 in 16 keys are passed into the gb_trees part instead of 1 in 4
* Immediate hibernate
Reasons for delay in hibernate were not clear.
Straight after creation the process will not be in receipt of messages (must wait for the manifest to be updated), so better to hibernate now. This also means the log PC023 provides more accurate information.
* Refactor BIC
This patch avoids the following:
- repeated replacement of the same element in the BIC (via get_kvrange), by checking presence via GET before sing SET
- Stops re-reading of all elements to discover high modified date
Also there appears to have been a bug where a missing HMD for the file is required to add to the cache. However, now the cache may be erased without erasing the HMD. This means that the cache can never be rebuilt
* Use correct size in test results
erts_debug:flat_size/1 returns size in words (i.e. 8 bytes on 64-bit CPU) not bytes
* Don't change summary record
As it is persisted as part of the file write, any change to the summary record cannot be rolled back
* Clerk to prompt L0 write
Simplifies the logic if the clerk request work for the penciller prompts L0 writes as well as Manifest changes.
The advantage now is that if the penciller memory is full, and PUT load stops, the clerk should still be able to prompt persistence. the penciller can therefore make use of dead time this way
* Add push on journal compact
If there has been a backlog, followed by a quiet period - there may be a large ledger cache left unpushed. Journal compaction events are about once per hour, so the performance overhead of a false push should be minimal, with the advantage of clearing any backlog before load starts again.
This is only relevant to riak users with very off/full batch type workloads.
* Extend tests
To more consistently trigger all overload scenarios
* Fix range keys smaller than prefix
Can't make end key an empty binary in this case, as it may be bigger than any keys within the range, but will appear to be smaller.
Unit tests and ct tests added to expose the potential issue
* Tidy-up
- Remove penciller logs which are no longer called
- Get pclerk to only wait MIN_TIMEOUT after doing work, in case there is a backlog
- Remove update_levelzero_cache function as it is unique to handle_call of push_mem, and simple enough to be inline
- Alight testutil slow offer with standard slow offer used
* Tidy-up
Remove pre-otp20 references.
Reinstate the check that the starting pid is still active, this was added to tidy up shutdown.
Resolve failure to run on otp20 due to `-if` sttaement
* Tidy up
Using null rather then {null, Key} is potentially clearer as it is not a concern what they Key is in this case, and removes a comparison step from the leveled_codec:endkey_passed/2 function.
There were issues with coverage in eunit tests as the leveled_pclerk shut down. This prompted a general tidy of leveled_pclerk (remove passing of LoopState into internal functions, and add dialyzer specs.
* Remove R16 relic
* Further testing another issue
The StartKey must always be less than or equal to the prefix when the first N characters are stripped, but this is not true of the EndKey (for the query) which does not have to be between the FirstKey and the LastKey.
If the EndKey query does not match it must be greater than the Prefix (as otherwise it would not have been greater than the FirstKey - so set to null.
* Fix unit test
Unit test had a typo - and result interpretation had a misunderstanding.
* Code and spec tidy
Also look to the cover the situation when the FirstKey is the same as the Prefix with tests.
This is, in theory, not an issue as it is the EndKey for each sublist which is indexed in leveled_tree. However, guard against it mapping to null here, just in case there are dangers lurking (note that tests will still pass without `M > N` guard in place.
* Hibernate on BIC complete
There are three situations when the BIC becomes complete:
- In a file created as part of a merge the BIS is learned in the merge
- After startup, files below L1 learn the block cache through reads that happen to read the block, eventually the while cache will be read, unless...
- Either before/after the cache is complete, it can get whiped by a timeout after a get_sqn request (e.g. as prompted by a journal compaction) ... it will then be re-filled of the back of get/get-range requests.
In all these situations we want to hibernate after the BIC is fill - to reflect the fact that the LoopState should now be relatively stable, so it is a good point to GC and rationalise location of data.
Previously on the the first base was covered. Now all three are covered through the bic_complete message.
* Test all index keys have same term
This works functionally, but is not optimised (the term is replicated in the index)
* Summaries with same index term
If the summary index all have the same index term - only the object keys need to be indexes
* Simplify case statements
We either match the pattern of <<Prefix:N, Suffix>> or the answer should be null
* OK for M == N
If M = N for the first key, it will have a suffix of <<>>. This will match (as expected) a query Start Key of the sam size, and be smaller than any query Start Key that has the same prefix.
If the query Start Key does not match the prefix - it will be null - as it must be smaller than the Prefix (as other wise the query Start Key would be bigger than the Last Key).
The constraint of M > N was introduced before the *_prefix_filter functions were checking the prefix, to avoid issues. Now the prefix is being checked, then M == N is ok.
* Simplify
Correct the test to use a binary field in the range.
To avoid further issue, only apply filter when everything is a binary() type.
* Add test for head_only mode
When leveled is used as a tictacaae key store (in parallel mode), the keys will be head_only entries. Double check they are handled as expected like object keys
* Revert previous change - must support typed buckets
Add assertion to confirm worthwhile optimisation
* Add support for configurable cache multiple (#375)
* Mas i370 patch e (#385)
Improvement to monitoring for efficiency and improved readability of logs and stats.
As part of this, where possible, tried to avoid updating loop state on READ messages in leveled processes (as was the case when tracking stats within each process).
No performance benefits found with change, but improved stats has helped discover other potential gains.
* Don't use fetch_cache below the page_cache level
* Don't time fetches due to SQN checks
SQN checks are all background processes
* Hibernate on SQN check
SQN check in the penciller is used for journal (all object) folds, but mainly for journal compaction. Use this to trigger hibernation where SST files stay quiet after the compaction check.
* Add catch for hibernate timeout
* Scale cache_size with level
Based on volume testing. Relatively speaking, far higher value to be gained from caches at higher levels (lower numbered levels). The cache at lower levels are proportionally much less efficient. so cache more at higher levels, where there is value, and less at lower levels where there is more cost relative to value.
* OTP 24 fix to cherry-pick
* Make minimal change to previous setup
Making significant change appears to not have had the expected positive improvement - so a more minimal change is proposed.
The assumption is that the cache only really gets used for double reads in the write path (e.g. where the application reads before a write) - and so a large cache make minimal difference, but no cache still has a downside.
* Introduce new types
* Mas i370 d30 sstmemory (#374)
* Don't time fetches due to SQN checks
SQN checks are all background processes
* Hibernate on SQN check
SQN check in the penciller is used for journal (all object) folds, but mainly for journal compaction. Use this to trigger hibernation where SST files stay quiet after the compaction check.
* Add catch for hibernate timeout
* Scale cache_size with level
Based on volume testing. Relatively speaking, far higher value to be gained from caches at higher levels (lower numbered levels). The cache at lower levels are proportionally much less efficient. so cache more at higher levels, where there is value, and less at lower levels where there is more cost relative to value.
* Make minimal change to previous setup
Making significant change appears to not have had the expected positive improvement - so a more minimal change is proposed.
The assumption is that the cache only really gets used for double reads in the write path (e.g. where the application reads before a write) - and so a large cache make minimal difference, but no cache still has a downside.
* Introduce new types
* More memory management
Clear blockindex_cache on timeout, and manually GC on pclerk after work.
* Add further garbage collection prompt
After fetching level zero, significant change in references in the penciller memory, so prompt a garbage_collect() at this point.
The L0 Pid has used a lot of memory in the construction of the file (something like 50MB). This won't be GC'd immediately. This is fine, as this will normally be short-lived. However if the SST file is switched levels ... then this may mean thta we have multiple SST files with memory not being GC'd.
Will not lead to immediate run time changes in SST or CDB logs. These log settings will only change once the new files are re-written.
To completely change the log level - a restart of the store is necessary with new startup options.
Previouslythe tinybloom was used within the SST file as an extra check to remove false fetches.
However the SST already has a low FPR check in the slot_index. If the newebloom was used (which is no longer per slot, but per sst), this can be shared with the penciller and then the penciller could use it and avoid the message pass.
the message pass may be blocked by a 2i query or a slot fetch request for a merge. So this should make performance within the Penciller snappier.
This is as a result of taking sst_timings within a volume test - where there was an average of + 100microsecs for each level that was dropped down. Given the bloom/slot checks were < 20 microsecs - there seems to be some further delay.
The bloom is a binary of > 64 bytes - so passing it around should not require a copy.
Compression can be switched between LZ4 and zlib (native).
The setting to determine if compression should happen on receipt is now a macro definition in leveled_codec.
Switch from magic hash to md5 - to hopefully remove the need for some
of the artificial jumps required to get expected fall positive ratios.
Also split the hash into two 16-bit integers. We assume that SegmentID
(from the perspective of AAE merkle/tictac trees) will always be at
least 16 bits. the idea is that hashes should be used in blooms and
indexes such that some advantage can be gained from just knowing the
segmentID - in particular when folding over all the keys in a bucket.
Performance testing has been difficult so far - I think due to “cloud”
mysteries.
The manifest and the logs are bloated by having the full file path for
every filename in there - given the root path is constant.
Could also cause issues if the mount point is ever changed.
As per comments though - if we auto-restart pclerk in the future this
will have to be re-considered.
Perhaps a re-starting pclerk should force some reset of this boolean on
startup perhaps by making a different work_for_clerk if in a virgin
state.
Need to add extra logging to understand why pclerk crashes in some
volume tests.
Penciller's Clerk <0.813.0> shutdown now complete for reason
{badarg,[{dict,fetch,[63,{dict,0,16,16,8,80,48,{[],[],[],[],[],[],[],[],[],[],[],[],[],[],[],[]},{{[],[],[],[],[],[],[],[],[],[],[],[],[],[],[],[]}}}],[{file,[100,105,99,116,46,101,114,108]},{line,126}]},{leveled_pclerk,handle_cast,2,[{file,[115,114,99,47,108,101,118,101,108,101,100,95,112,99,108,101,114,107,46,101,114,108]},{line,96}]},{gen_server,handle_msg,5,[{file,[103,101,110,95,115,101,114,118,101,114,46,101,114,108]},{line,604}]},{proc_lib,init_p_do_apply,3,[{file,[112,114,111,99,95,108,105,98,46,101,114,108]},{line,239}]}]}
Should only be prompted after prompt deletions had bene updated.
Perhaps a race whereby somehow it is prompted again after it is emptied.
This has refactored code with the implementation of the manifest
isolated in to a seperate module, and the pure async relationship
between penciller and their clerk. However, the manifest is just a
simple list at each level.
correct the previous change to make sure that deletes are not confirmed
when work is outstanding, but also make sure that the clerk only ever
casts the Penciller so no deadlocks can ensue.
Going to abandond this branch for now. The change is beoming
excessively time consuming, and it is not clear that a smaller change
might not achieve more of the objectives.
All this is broken - but perhaps could get picke dup another day.
This is desirable to add back in going forward, but wasn't implemented
in a safe or clear way.
The way the bloom was or was not on the LoopState was clumsy, and it got
persisted in multiple places without a CRC check.
Intention to implement back in wherby it is requested on-demand by the
Penciller, and then the SFT worker lifts it off disk and CRC checks it.
So it is never on the SFT LoopState. Also it will be easier to control
the logic over which levels have the bloom in the Penciller.
