TimeRangeTracker as point of contention when many threads reading a StoreFile
Fixes HBASE-16074 ITBLL fails, reports lost big or tiny families broken
scanning because of a side effect of a clean up in HBASE-15650 to make
TimeRange construction consistent exposed a latent issue in
TimeRange#compare. See HBASE-16074 for more detail.
Also change HFile Writer constructor so we pass in the TimeRangeTracker, if one,
on construction rather than set later (the flag and reference were not volatile
so could have made for issues in concurrent case). And make sure the construction
of a TimeRange from a TimeRangeTracer on open of an HFile Reader never makes a
bad minimum value, one that would preclude us reading any values from a file
(set min to 0)
M hbase-common/src/main/java/org/apache/hadoop/hbase/io/TimeRange.java
Call through to next constructor (if minStamp was 0, we'd skip setting
allTime=true). Add asserts that timestamps are not < 0 cos it messes
us up if they are (we already were checking for < 0 on construction but
assert passed in timestamps are not < 0).
M hbase-server/src/main/java/org/apache/hadoop/hbase/regionserver/HStore.java
Add constructor override that takes a TimeRangeTracker (set when flushing
but not when compacting)
M hbase-server/src/main/java/org/apache/hadoop/hbase/regionserver/Store.java
Add override creating an HFile in tmp that takes a TimeRangeTracker
M hbase-server/src/main/java/org/apache/hadoop/hbase/regionserver/StoreFile.java
Add override for HFile Writer that takes a TimeRangeTracker Take it on
construction instead of having it passed by a setter later (flags and
reference set by the setter were not volatile... could have been prob
in concurrent case)
M hbase-server/src/main/java/org/apache/hadoop/hbase/regionserver/TimeRangeTracker.java
Log WARN if bad initial TimeRange value (and then 'fix' it)
M hbase-server/src/test/java/org/apache/hadoop/hbase/regionserver/TestTimeRangeTracker.java
A few tests to prove serialization works as expected and that we'll get a bad min if not constructed properly.
M hbase-server/src/main/java/org/apache/hadoop/hbase/regionserver/ScanQueryMatcher.java
Handle OLDEST_TIMESTAMP explictly. Don't expect TimeRange to do it.
M hbase-server/src/test/java/org/apache/hadoop/hbase/regionserver/TestQueryMatcher.java
Refactor from junit3 to junit4 and add test for this weird case.
Refactor so we use the immutable, unsynchronized TimeRange when doing
time-based checks at read time rather than use heavily synchronized
TimeRangeTracker; let TimeRangeTracker be for write-time only.
While in here, changed the Segment stuff so that when an immutable
segment, it uses TimeRange rather than TimeRangeTracker too.
M hbase-common/src/main/java/org/apache/hadoop/hbase/io/TimeRange.java
Make allTime final.
Add a includesTimeRange method copied from TimeRangeTracker.
M hbase-server/src/main/java/org/apache/hadoop/hbase/regionserver/TimeRangeTracker.java
Change name of a few methods so they match TimeRange methods that do
same thing.
(getTimeRangeTracker, getTimeRange, toTimeRange) add utility methods
M hbase-server/src/main/java/org/apache/hadoop/hbase/regionserver/ImmutableSegment.java
Change ImmutableSegment so it uses a TimeRange rather than
TimeRangeTracker.. it is read-only. Redo shouldSeek, getMinTimestamp,
updateMetaInfo, and getTimeRangeTracker so we use TimeRange-based
implementations instead of TimeRangeTracker implementations.
M hbase-server/src/main/java/org/apache/hadoop/hbase/regionserver/MutableSegment.java
Implement shouldSeek, getMinTimestamp, updateMetaInfo, and
getTimeRangeTracker using TimeRangeTracker.
M hbase-server/src/main/java/org/apache/hadoop/hbase/regionserver/Segment.java
Make methods that were using TimeRangeTracker abstract and instead
have the implementations do these methods how they want either using
TimeRangeTracker when a mutable segment or TimeRange when an immutable
segment.
M hbase-server/src/main/java/org/apache/hadoop/hbase/regionserver/StoreFile.java
Change Reader to use TimeRange-based checks instead of
TimeRangeTracker.
Signed-off-by: stack <stack@apache.org>
M hbase-server/src/main/java/org/apache/hadoop/hbase/regionserver/ScanQueryMatcher.java
moreRowsMayExistAfterCell Exploit the fact a Scan is a Get Scan. Also save compares
if no non-default stopRow.
M hbase-server/src/main/java/org/apache/hadoop/hbase/regionserver/StoreScanner.java
optimize Add doc on what is being optimized. Also, if a Get Scan, do not
optimize else we'll keep going after our row is DONE.
Another place to make use of the Get Scan fact is when we are DONE.. if
Get Scan, we can close out the scan.
M hbase-server/src/test/java/org/apache/hadoop/hbase/regionserver/TestStoreScanner.java
Add tests for Get Scans and optimize around block loading.
When we read from HDFS, we overread to pick up the next blocks header.
Doing this saves a seek as we move through the hfile; we save having to
do an explicit seek just to read the block header every time we need to
read the body. We used to read in the next header as part of the
current blocks buffer. This buffer was then what got persisted to
blockcache; so we were over-persisting: our block plus the next blocks'
header (33 bytes).
This patch undoes this over-persisting.
Removes support for version 1 blocks (0.2 was added in hbase-0.92.0).
Not needed any more.
There is an open question on whether checksums should be persisted
when caching. The code seems to say no but if cache is SSD backed or
backed by anything that does not do error correction, we'll want
checksums.
Adds loads of documentation.
M hbase-common/src/main/java/org/apache/hadoop/hbase/io/hfile/BlockType.java
(write) Add writing from a ByteBuff.
M hbase-common/src/main/java/org/apache/hadoop/hbase/nio/ByteBuff.java
(toString) Add one so ByteBuff looks like ByteBuffer when you click on
it in IDE
M hbase-server/src/main/java/org/apache/hadoop/hbase/io/hfile/HFileBlock.java
Remove support for version 1 blocks.
Cleaned up handling of metadata added when we serialize a block to
caches. Metadata is smaller now.
When we serialize (used when caching), do not persist the next blocks
header if present.
Removed a bunch of methods, a few of which had overlapping
functionality and others that exposed too much of our internals.
Also removed a bunch of constructors and unified the constructors we
had left over making them share a common init method.
Shutdown access to defines that should only be used internally here.
Renamed all to do w/ 'EXTRA' and 'extraSerialization' to instead talk
about metadata saved to caches; was unclear previously what EXTRA was
about.
Renamed static final declarations as all uppercase.
(readBlockDataInternal): Redid. Couldn't make sense of it previously.
Undid heavy-duty parse of header by constructing HFileBlock. Other
cleanups. Its 1/3rd the length it used to be. More to do in here.
When we read from HDFS, we overread to pick up the next blocks header.
Doing this saves a seek as we move through the hfile; we save having to
do an explicit seek just to read the block header every time we need to
read the body. We used to read in the next header as part of the
current blocks buffer. This buffer was then what got persisted to
blockcache; so we were over-persisting: our block plus the next blocks'
header (33 bytes).
This patch undoes this over-persisting.
Removes support for version 1 blocks (0.2 was added in hbase-0.92.0).
Not needed any more.
There is an open question on whether checksums should be persisted
when caching. The code seems to say no but if cache is SSD backed or
backed by anything that does not do error correction, we'll want
checksums.
Adds loads of documentation.
M hbase-common/src/main/java/org/apache/hadoop/hbase/io/hfile/BlockType.java
(write) Add writing from a ByteBuff.
M hbase-common/src/main/java/org/apache/hadoop/hbase/nio/ByteBuff.java
(toString) Add one so ByteBuff looks like ByteBuffer when you click on
it in IDE
M hbase-server/src/main/java/org/apache/hadoop/hbase/io/hfile/HFileBlock.java
Remove support for version 1 blocks.
Cleaned up handling of metadata added when we serialize a block to
caches. Metadata is smaller now.
When we serialize (used when caching), do not persist the next blocks
header if present.
Removed a bunch of methods, a few of which had overlapping
functionality and others that exposed too much of our internals.
Also removed a bunch of constructors and unified the constructors we
had left over making them share a common init method.
Shutdown access to defines that should only be used internally here.
Renamed all to do w/ 'EXTRA' and 'extraSerialization' to instead talk
about metadata saved to caches; was unclear previously what EXTRA was
about.
Renamed static final declarations as all uppercase.
(readBlockDataInternal): Redid. Couldn't make sense of it previously.
Undid heavy-duty parse of header by constructing HFileBlock. Other
cleanups. Its 1/3rd the length it used to be. More to do in here.
M hbase-server/src/main/java/org/apache/hadoop/hbase/regionserver/StoreScanner.java
optimize Add doc on what is being optimized. Also, if a Get Scan, do not
optimize else we'll keep going after our row is DONE.
Another place to make use of the Get Scan fact is when we are DONE.. if
Get Scan, we can close out the scan.
M hbase-server/src/test/java/org/apache/hadoop/hbase/regionserver/TestStoreScanner.java
Add tests for Get Scans and optimize around block loading.
M hbase-common/src/main/java/org/apache/hadoop/hbase/io/hfile/HFileContext.java
Make it emit its toString in format that matches the way we log
elsewhere
M hbase-server/src/main/java/org/apache/hadoop/hbase/io/hfile/HFile.java
Capitalize statics.
M hbase-server/src/main/java/org/apache/hadoop/hbase/io/hfile/HFileBlock.java
Verify and cleanup documentation of hfileblock format at head of class.
Explain what 'EXTRA_SERIALIZATION_SPACE' is all about.
Connect how we serialize and deserialize... done in different places
and one way when pulling from HDFS and another when pulling from cache
(TO BE FIXED). Shut down a load of public access.
M hbase-server/src/main/java/org/apache/hadoop/hbase/io/hfile/HFileReaderImpl.java
Add trace-level logging
M hbase-server/src/main/java/org/apache/hadoop/hbase/io/hfile/HFileScanner.java
Make it Closeable
Summary:
Use less contended things for metrics.
For histogram which was the largest culprit we use FastLongHistogram
For atomic long where possible we now use counter.
Test Plan: unit tests
Reviewers:
Subscribers:
Differential Revision: https://reviews.facebook.net/D54381
zhangduo
stack
anoopsjohn
Matteo Bertozzi
Jerry He
Andrew Purtell
tedyu
chenheng
Elliott Clark
Ashu Pachauri
Enis Soztutar
Ramkrishna
Lars Hofhansl
Apekshit
Youngjoon Kim
Sean Mackrory
Sean Busbey
Jonathan M Hsieh
Junegunn Choi
Phil Yang
Stephen Yuan Jiang
Mikhail Antonov
eshcar
ramkrishna
Misty Stanley-Jones
Gary Helmling