You are viewing a plain text version of this content. The canonical link for it is here.
Posted to github@arrow.apache.org by GitBox <gi...@apache.org> on 2022/10/25 23:10:33 UTC
[GitHub] [arrow] fatemehp opened a new pull request, #14509: PARQUET-2204: [parquet-cpp] TypedColumnReaderImpl::Skip should reuse scratch space
fatemehp opened a new pull request, #14509:
URL: https://github.com/apache/arrow/pull/14509
TypedColumnReaderImpl::Skip allocates scratch space on every call. The scratch space is used to read rep/def levels and values and throw them away. The memory allocation slows down the skip based on microbenchmarks. The scratch space can be allocated once and re-used.
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: github-unsubscribe@arrow.apache.org
For queries about this service, please contact Infrastructure at:
users@infra.apache.org
[GitHub] [arrow] emkornfield commented on pull request #14509: PARQUET-2204: [parquet-cpp] TypedColumnReaderImpl::Skip should reuse scratch space
Posted by GitBox <gi...@apache.org>.
emkornfield commented on PR #14509:
URL: https://github.com/apache/arrow/pull/14509#issuecomment-1297655387
> We need to consider the memory implications of this change. This means that if at least one Skip is requested, the scratch space will be allocated on the heap and kept until the column reader is destroyed. The scratch space can be as big as 12 KB. If we have 1024 column readers open at one time, that means a 12 MB overhead. Is it common to have this many readers open at the same time? If yes, is this overhead acceptable?
That seems OK to me, another option would be to read the feature flag. This breaks only after a second call to skip?
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: github-unsubscribe@arrow.apache.org
For queries about this service, please contact Infrastructure at:
users@infra.apache.org
[GitHub] [arrow] fatemehp commented on pull request #14509: PARQUET-2204: [parquet-cpp] TypedColumnReaderImpl::Skip should reuse scratch space
Posted by GitBox <gi...@apache.org>.
fatemehp commented on PR #14509:
URL: https://github.com/apache/arrow/pull/14509#issuecomment-1291242278
I am planning to check in the microbenchmark as a separate pull request.
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: github-unsubscribe@arrow.apache.org
For queries about this service, please contact Infrastructure at:
users@infra.apache.org
[GitHub] [arrow] emkornfield commented on pull request #14509: PARQUET-2204: [parquet-cpp] TypedColumnReaderImpl::Skip should reuse scratch space
Posted by GitBox <gi...@apache.org>.
emkornfield commented on PR #14509:
URL: https://github.com/apache/arrow/pull/14509#issuecomment-1292543272
Looks like this change is the same as in the new PR with microbenchmark, can we close in favor of that?
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: github-unsubscribe@arrow.apache.org
For queries about this service, please contact Infrastructure at:
users@infra.apache.org
[GitHub] [arrow] github-actions[bot] commented on pull request #14509: PARQUET-2204: [parquet-cpp] TypedColumnReaderImpl::Skip should reuse scratch space
Posted by GitBox <gi...@apache.org>.
github-actions[bot] commented on PR #14509:
URL: https://github.com/apache/arrow/pull/14509#issuecomment-1291236975
:warning: Ticket **has no components in JIRA**, make sure you assign one.
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: github-unsubscribe@arrow.apache.org
For queries about this service, please contact Infrastructure at:
users@infra.apache.org
[GitHub] [arrow] pitrou commented on a diff in pull request #14509: PARQUET-2204: [parquet-cpp] TypedColumnReaderImpl::Skip should reuse scratch space
Posted by GitBox <gi...@apache.org>.
pitrou commented on code in PR #14509:
URL: https://github.com/apache/arrow/pull/14509#discussion_r1023010447
##########
cpp/src/parquet/column_reader.cc:
##########
@@ -1151,6 +1159,14 @@ int64_t TypedColumnReaderImpl<DType>::ReadBatchSpaced(
return total_values;
}
+template <typename DType>
+void TypedColumnReaderImpl<DType>::InitScratchForSkip(int64_t batch_size) {
Review Comment:
What if the batch size is not the same between calls?
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: github-unsubscribe@arrow.apache.org
For queries about this service, please contact Infrastructure at:
users@infra.apache.org
[GitHub] [arrow] fatemehp commented on a diff in pull request #14509: PARQUET-2204: [parquet-cpp] TypedColumnReaderImpl::Skip should reuse scratch space
Posted by GitBox <gi...@apache.org>.
fatemehp commented on code in PR #14509:
URL: https://github.com/apache/arrow/pull/14509#discussion_r1009849826
##########
cpp/src/parquet/column_reader.cc:
##########
@@ -928,6 +928,14 @@ class TypedColumnReaderImpl : public TypedColumnReader<DType>,
this->exposed_encoding_ = encoding;
}
+ // Allocate enough scratch space to accommodate skipping 16-bit levels or any
+ // value type for batch_size.
+ void InitScratchForSkip(int64_t batch_size);
+
+ // Scrtach space for reading and throwing away rep/def levels and values when
Review Comment:
Done.
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: github-unsubscribe@arrow.apache.org
For queries about this service, please contact Infrastructure at:
users@infra.apache.org
[GitHub] [arrow] fatemehp commented on pull request #14509: PARQUET-2204: [parquet-cpp] TypedColumnReaderImpl::Skip should reuse scratch space
Posted by GitBox <gi...@apache.org>.
fatemehp commented on PR #14509:
URL: https://github.com/apache/arrow/pull/14509#issuecomment-1292675017
I removed these commits from the benchmark pull request.
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: github-unsubscribe@arrow.apache.org
For queries about this service, please contact Infrastructure at:
users@infra.apache.org
[GitHub] [arrow] github-actions[bot] commented on pull request #14509: PARQUET-2204: [parquet-cpp] TypedColumnReaderImpl::Skip should reuse scratch space
Posted by GitBox <gi...@apache.org>.
github-actions[bot] commented on PR #14509:
URL: https://github.com/apache/arrow/pull/14509#issuecomment-1291236993
:warning: Ticket **has not been started in JIRA**, please click 'Start Progress'.
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: github-unsubscribe@arrow.apache.org
For queries about this service, please contact Infrastructure at:
users@infra.apache.org
[GitHub] [arrow] fatemehp commented on pull request #14509: PARQUET-2204: [parquet-cpp] TypedColumnReaderImpl::Skip should reuse scratch space
Posted by GitBox <gi...@apache.org>.
fatemehp commented on PR #14509:
URL: https://github.com/apache/arrow/pull/14509#issuecomment-1291234194
@emkornfield could you take a look at this pull request? Thanks!
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: github-unsubscribe@arrow.apache.org
For queries about this service, please contact Infrastructure at:
users@infra.apache.org
[GitHub] [arrow] pitrou merged pull request #14509: PARQUET-2204: [parquet-cpp] TypedColumnReaderImpl::Skip should reuse scratch space
Posted by GitBox <gi...@apache.org>.
pitrou merged PR #14509:
URL: https://github.com/apache/arrow/pull/14509
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: github-unsubscribe@arrow.apache.org
For queries about this service, please contact Infrastructure at:
users@infra.apache.org
[GitHub] [arrow] emkornfield commented on a diff in pull request #14509: PARQUET-2204: [parquet-cpp] TypedColumnReaderImpl::Skip should reuse scratch space
Posted by GitBox <gi...@apache.org>.
emkornfield commented on code in PR #14509:
URL: https://github.com/apache/arrow/pull/14509#discussion_r1009841780
##########
cpp/src/parquet/column_reader.cc:
##########
@@ -928,6 +928,14 @@ class TypedColumnReaderImpl : public TypedColumnReader<DType>,
this->exposed_encoding_ = encoding;
}
+ // Allocate enough scratch space to accommodate skipping 16-bit levels or any
+ // value type for batch_size.
+ void InitScratchForSkip(int64_t batch_size);
+
+ // Scrtach space for reading and throwing away rep/def levels and values when
Review Comment:
```suggestion
// Scratch space for reading and throwing away rep/def levels and values when
```
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: github-unsubscribe@arrow.apache.org
For queries about this service, please contact Infrastructure at:
users@infra.apache.org
[GitHub] [arrow] fatemehp commented on pull request #14509: PARQUET-2204: [parquet-cpp] TypedColumnReaderImpl::Skip should reuse scratch space
Posted by GitBox <gi...@apache.org>.
fatemehp commented on PR #14509:
URL: https://github.com/apache/arrow/pull/14509#issuecomment-1295282796
We need to consider the memory implications of this change. This means that if at least one Skip is requested, the scratch space will be allocated on the heap and kept until the column reader is destroyed. The scratch space can be as big as 12 KB. If we have 1024 column readers open at one time, that means a 12 MB overhead. Is it common to have this many readers open at the same time? If yes, is this overhead acceptable?
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: github-unsubscribe@arrow.apache.org
For queries about this service, please contact Infrastructure at:
users@infra.apache.org
[GitHub] [arrow] fatemehp commented on pull request #14509: PARQUET-2204: [parquet-cpp] TypedColumnReaderImpl::Skip should reuse scratch space
Posted by GitBox <gi...@apache.org>.
fatemehp commented on PR #14509:
URL: https://github.com/apache/arrow/pull/14509#issuecomment-1341596024
@pitrou, @emkornfield I have pulled in the latest changes, and updated the code so that the RecordReader uses the same scratch space. Please take a look.
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: github-unsubscribe@arrow.apache.org
For queries about this service, please contact Infrastructure at:
users@infra.apache.org
[GitHub] [arrow] fatemehp commented on pull request #14509: PARQUET-2204: [parquet-cpp] TypedColumnReaderImpl::Skip should reuse scratch space
Posted by GitBox <gi...@apache.org>.
fatemehp commented on PR #14509:
URL: https://github.com/apache/arrow/pull/14509#issuecomment-1298825412
@emkornfield Could you give some context about when we normally use the feature flags? If we want to control this using a flag, that means we will support both cases of 1) allocating within the Skip function and 2) in the column reader. I am wondering if that is doing more than what we need.
Also, I don't fully understand your question here: "This breaks only after a second call to skip?"
The scratch space is allocated on the "first" call to skip and will be retained until the column reader is destroyed.
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: github-unsubscribe@arrow.apache.org
For queries about this service, please contact Infrastructure at:
users@infra.apache.org
[GitHub] [arrow] emkornfield commented on pull request #14509: PARQUET-2204: [parquet-cpp] TypedColumnReaderImpl::Skip should reuse scratch space
Posted by GitBox <gi...@apache.org>.
emkornfield commented on PR #14509:
URL: https://github.com/apache/arrow/pull/14509#issuecomment-1331174859
> @emkornfield Could you give some context about when we normally use the feature flags? If we want to control this using a flag, that means we will support both cases of 1) allocating within the Skip function and 2) in the column reader. I am wondering if that is doing more than what we need.
> Also, I don't fully understand your question here: "This breaks only after a second call to skip?"
The scratch space is allocated on the "first" call to skip and will be retained until the column reader is destroyed.
sorry I think this answers the question, I think I meant help instead of breaks.
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: github-unsubscribe@arrow.apache.org
For queries about this service, please contact Infrastructure at:
users@infra.apache.org
[GitHub] [arrow] pitrou commented on pull request #14509: PARQUET-2204: [parquet-cpp] TypedColumnReaderImpl::Skip should reuse scratch space
Posted by GitBox <gi...@apache.org>.
pitrou commented on PR #14509:
URL: https://github.com/apache/arrow/pull/14509#issuecomment-1315566656
Can you merge the latest changes from git master?
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: github-unsubscribe@arrow.apache.org
For queries about this service, please contact Infrastructure at:
users@infra.apache.org
[GitHub] [arrow] fatemehp commented on a diff in pull request #14509: PARQUET-2204: [parquet-cpp] TypedColumnReaderImpl::Skip should reuse scratch space
Posted by GitBox <gi...@apache.org>.
fatemehp commented on code in PR #14509:
URL: https://github.com/apache/arrow/pull/14509#discussion_r1042681848
##########
cpp/src/parquet/column_reader.cc:
##########
@@ -1151,6 +1159,14 @@ int64_t TypedColumnReaderImpl<DType>::ReadBatchSpaced(
return total_values;
}
+template <typename DType>
+void TypedColumnReaderImpl<DType>::InitScratchForSkip(int64_t batch_size) {
Review Comment:
I removed this argument and clarified this in the code. The batch size is constant and will not change.
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: github-unsubscribe@arrow.apache.org
For queries about this service, please contact Infrastructure at:
users@infra.apache.org
[GitHub] [arrow] pitrou commented on pull request #14509: PARQUET-2204: [parquet-cpp] TypedColumnReaderImpl::Skip should reuse scratch space
Posted by GitBox <gi...@apache.org>.
pitrou commented on PR #14509:
URL: https://github.com/apache/arrow/pull/14509#issuecomment-1342905502
I get the following benchmark numbers:
* before:
```
-------------------------------------------------------------------------------------------------------------------
Benchmark Time CPU Iterations UserCounters...
-------------------------------------------------------------------------------------------------------------------
ColumnReaderSkipInt32/Repetition:0/BatchSize:100 2256949 ns 2256574 ns 310 bytes_per_second=2.1131G/s
ColumnReaderSkipInt32/Repetition:0/BatchSize:1000 322274 ns 322302 ns 2147 bytes_per_second=14.7947G/s
ColumnReaderSkipInt32/Repetition:0/BatchSize:10000 123075 ns 123135 ns 5695 bytes_per_second=38.7247G/s
ColumnReaderSkipInt32/Repetition:0/BatchSize:100000 34654 ns 34707 ns 19814 bytes_per_second=137.388G/s
ColumnReaderSkipInt32/Repetition:1/BatchSize:100 4483059 ns 4482271 ns 159 bytes_per_second=579.976M/s
ColumnReaderSkipInt32/Repetition:1/BatchSize:1000 974701 ns 974606 ns 719 bytes_per_second=2.60483G/s
ColumnReaderSkipInt32/Repetition:1/BatchSize:10000 630965 ns 630913 ns 1111 bytes_per_second=4.02382G/s
ColumnReaderSkipInt32/Repetition:1/BatchSize:100000 191333 ns 191349 ns 3553 bytes_per_second=13.2672G/s
ColumnReaderSkipInt32/Repetition:2/BatchSize:100 5919031 ns 5917853 ns 119 bytes_per_second=465.768M/s
ColumnReaderSkipInt32/Repetition:2/BatchSize:1000 1422720 ns 1422501 ns 488 bytes_per_second=1.89226G/s
ColumnReaderSkipInt32/Repetition:2/BatchSize:10000 1008342 ns 1008178 ns 698 bytes_per_second=2.66991G/s
ColumnReaderSkipInt32/Repetition:2/BatchSize:100000 306069 ns 306087 ns 2281 bytes_per_second=8.79405G/s
```
* after:
```
ColumnReaderSkipInt32/Repetition:0/BatchSize:100 246589 ns 246512 ns 2869 bytes_per_second=19.3434G/s
ColumnReaderSkipInt32/Repetition:0/BatchSize:1000 114571 ns 114636 ns 6122 bytes_per_second=41.5957G/s
ColumnReaderSkipInt32/Repetition:0/BatchSize:10000 98790 ns 98844 ns 7101 bytes_per_second=48.2414G/s
ColumnReaderSkipInt32/Repetition:0/BatchSize:100000 32585 ns 32662 ns 22074 bytes_per_second=145.992G/s
ColumnReaderSkipInt32/Repetition:1/BatchSize:100 1716748 ns 1716514 ns 408 bytes_per_second=1.47897G/s
ColumnReaderSkipInt32/Repetition:1/BatchSize:1000 644526 ns 644500 ns 1091 bytes_per_second=3.93899G/s
ColumnReaderSkipInt32/Repetition:1/BatchSize:10000 567605 ns 567578 ns 1241 bytes_per_second=4.47283G/s
ColumnReaderSkipInt32/Repetition:1/BatchSize:100000 180075 ns 180121 ns 3862 bytes_per_second=14.0943G/s
ColumnReaderSkipInt32/Repetition:2/BatchSize:100 2880737 ns 2880209 ns 243 bytes_per_second=956.995M/s
ColumnReaderSkipInt32/Repetition:2/BatchSize:1000 1040618 ns 1040539 ns 676 bytes_per_second=2.58687G/s
ColumnReaderSkipInt32/Repetition:2/BatchSize:10000 916319 ns 916171 ns 767 bytes_per_second=2.93803G/s
ColumnReaderSkipInt32/Repetition:2/BatchSize:100000 288829 ns 288867 ns 2418 bytes_per_second=9.31826G/s
```
Nice improvement!
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: github-unsubscribe@arrow.apache.org
For queries about this service, please contact Infrastructure at:
users@infra.apache.org
[GitHub] [arrow] fatemehp commented on pull request #14509: PARQUET-2204: [parquet-cpp] TypedColumnReaderImpl::Skip should reuse scratch space
Posted by GitBox <gi...@apache.org>.
fatemehp commented on PR #14509:
URL: https://github.com/apache/arrow/pull/14509#issuecomment-1292687122
Here are benchmark results before and after the change proposed in this pull request. Only including the Skip since this change does not affect the read performance. We get up to 15X reduction when the batch size (last parameter) is 1, which means we are repeatedly re-allocating the scratch space.
```
PARAMETERS:
1) definition level
2) repetition level
3) 1 for Skip, 0 for ReadBatch (here we only have Skip)
4) batch size
BEFORE
-------------------------------------------------------------------------------
Benchmark Time CPU Iterations
-------------------------------------------------------------------------------
BM_Skip/0/0/1/1/iterations:1000 150319842 ns 149567587 ns 1000
BM_Skip/0/0/1/1000/iterations:1000 244565 ns 244931 ns 1000
BM_Skip/0/0/1/10000/iterations:1000 115395 ns 115924 ns 1000
BM_Skip/0/0/1/100000/iterations:1000 115241 ns 115916 ns 1000
BM_Skip/1/0/1/1/iterations:1000 149224507 ns 148683644 ns 1000
BM_Skip/1/0/1/1000/iterations:1000 805812 ns 805417 ns 1000
BM_Skip/1/0/1/10000/iterations:1000 702999 ns 700108 ns 1000
BM_Skip/1/0/1/100000/iterations:1000 654163 ns 651947 ns 1000
BM_Skip/1/1/1/1/iterations:1000 165600118 ns 164864530 ns 1000
BM_Skip/1/1/1/1000/iterations:1000 1130975 ns 1130252 ns 1000
BM_Skip/1/1/1/10000/iterations:1000 1009628 ns 1009589 ns 1000
BM_Skip/1/1/1/100000/iterations:1000 1029064 ns 1028726 ns 1000
AFTER
-------------------------------------------------------------------------------
Benchmark Time CPU Iterations
-------------------------------------------------------------------------------
BM_Skip/0/0/1/1/iterations:1000 10280337 ns 10234495 ns 1000
BM_Skip/0/0/1/1000/iterations:1000 101228 ns 101436 ns 1000
BM_Skip/0/0/1/10000/iterations:1000 96565 ns 96648 ns 1000
BM_Skip/0/0/1/100000/iterations:1000 96598 ns 96814 ns 1000
BM_Skip/1/0/1/1/iterations:1000 25771605 ns 25718891 ns 1000
BM_Skip/1/0/1/1000/iterations:1000 651609 ns 650940 ns 1000
BM_Skip/1/0/1/10000/iterations:1000 660890 ns 654217 ns 1000
BM_Skip/1/0/1/100000/iterations:1000 640417 ns 636855 ns 1000
BM_Skip/1/1/1/1/iterations:1000 36639124 ns 36433537 ns 1000
BM_Skip/1/1/1/1000/iterations:1000 978058 ns 976403 ns 1000
BM_Skip/1/1/1/10000/iterations:1000 997193 ns 996529 ns 1000
BM_Skip/1/1/1/100000/iterations:1000 999080 ns 993296 ns 1000
```
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: github-unsubscribe@arrow.apache.org
For queries about this service, please contact Infrastructure at:
users@infra.apache.org
[GitHub] [arrow] github-actions[bot] commented on pull request #14509: PARQUET-2204: [parquet-cpp] TypedColumnReaderImpl::Skip should reuse scratch space
Posted by GitBox <gi...@apache.org>.
github-actions[bot] commented on PR #14509:
URL: https://github.com/apache/arrow/pull/14509#issuecomment-1291236949
https://issues.apache.org/jira/browse/PARQUET-2204
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: github-unsubscribe@arrow.apache.org
For queries about this service, please contact Infrastructure at:
users@infra.apache.org