sapling/tests/test-clone-uncompressed.t

93 lines
2.7 KiB
Perl
Raw Normal View History

streamclone: use backgroundfilecloser (issue4889) Closing files that have been appended to is slow on Windows/NTFS. CloseHandle() calls on this platform often take 1-10ms - and that's on my i7-6700K Skylake processor with a modern and fast SSD. Contrast with other I/O operations, such as writing data, which take <100us. This means that creating/appending thousands of files can add significant overhead. For example, cloning mozilla-central creates ~232,000 revlog files. Assuming 1ms per CloseHandle(), that yields 232s (3:52) of wall time waiting for file closes! The impact of this overhead can be measured most directly when applying stream clone bundles. Applying these files is effectively uncompressing a tar archive (read: it's very fast). Using a RAM disk (read: no I/O wait), the difference in wall time for a `hg debugapplystreamclonebundle` for a ~1731 MB mozilla-central bundle between Windows and Linux from the same machine is drastic: Linux: ~12.8s (128MB/s) Windows: ~352.0s (4.7MB/s) Windows is ~27.5x slower. Yikes! After this patch: Linux: ~12.8s (128MB/s) Windows: ~102.1s (16.1MB/s) Windows is now ~3.4x faster. Unfortunately, it is still ~8x slower than Linux. Profiling reveals a few hot code paths that could likely be improved. But those are for other patches. This patch introduces test-clone-uncompressed.t because existing tests of `clone --uncompressed` are scattered about and adding a variation for background thread closing to e.g. test-http.t doesn't feel correct.
2016-01-15 00:44:01 +03:00
#require serve
Initialize repository
the status call is to check for issue5130
streamclone: use backgroundfilecloser (issue4889) Closing files that have been appended to is slow on Windows/NTFS. CloseHandle() calls on this platform often take 1-10ms - and that's on my i7-6700K Skylake processor with a modern and fast SSD. Contrast with other I/O operations, such as writing data, which take <100us. This means that creating/appending thousands of files can add significant overhead. For example, cloning mozilla-central creates ~232,000 revlog files. Assuming 1ms per CloseHandle(), that yields 232s (3:52) of wall time waiting for file closes! The impact of this overhead can be measured most directly when applying stream clone bundles. Applying these files is effectively uncompressing a tar archive (read: it's very fast). Using a RAM disk (read: no I/O wait), the difference in wall time for a `hg debugapplystreamclonebundle` for a ~1731 MB mozilla-central bundle between Windows and Linux from the same machine is drastic: Linux: ~12.8s (128MB/s) Windows: ~352.0s (4.7MB/s) Windows is ~27.5x slower. Yikes! After this patch: Linux: ~12.8s (128MB/s) Windows: ~102.1s (16.1MB/s) Windows is now ~3.4x faster. Unfortunately, it is still ~8x slower than Linux. Profiling reveals a few hot code paths that could likely be improved. But those are for other patches. This patch introduces test-clone-uncompressed.t because existing tests of `clone --uncompressed` are scattered about and adding a variation for background thread closing to e.g. test-http.t doesn't feel correct.
2016-01-15 00:44:01 +03:00
$ hg init server
$ cd server
$ touch foo
$ hg -q commit -A -m initial
>>> for i in range(1024):
... with open(str(i), 'wb') as fh:
... fh.write(str(i))
$ hg -q commit -A -m 'add a lot of files'
$ hg st
streamclone: use backgroundfilecloser (issue4889) Closing files that have been appended to is slow on Windows/NTFS. CloseHandle() calls on this platform often take 1-10ms - and that's on my i7-6700K Skylake processor with a modern and fast SSD. Contrast with other I/O operations, such as writing data, which take <100us. This means that creating/appending thousands of files can add significant overhead. For example, cloning mozilla-central creates ~232,000 revlog files. Assuming 1ms per CloseHandle(), that yields 232s (3:52) of wall time waiting for file closes! The impact of this overhead can be measured most directly when applying stream clone bundles. Applying these files is effectively uncompressing a tar archive (read: it's very fast). Using a RAM disk (read: no I/O wait), the difference in wall time for a `hg debugapplystreamclonebundle` for a ~1731 MB mozilla-central bundle between Windows and Linux from the same machine is drastic: Linux: ~12.8s (128MB/s) Windows: ~352.0s (4.7MB/s) Windows is ~27.5x slower. Yikes! After this patch: Linux: ~12.8s (128MB/s) Windows: ~102.1s (16.1MB/s) Windows is now ~3.4x faster. Unfortunately, it is still ~8x slower than Linux. Profiling reveals a few hot code paths that could likely be improved. But those are for other patches. This patch introduces test-clone-uncompressed.t because existing tests of `clone --uncompressed` are scattered about and adding a variation for background thread closing to e.g. test-http.t doesn't feel correct.
2016-01-15 00:44:01 +03:00
$ hg serve -p $HGPORT -d --pid-file=hg.pid
$ cat hg.pid >> $DAEMON_PIDS
$ cd ..
Basic clone
$ hg clone --uncompressed -U http://localhost:$HGPORT clone1
streaming all changes
1027 files to transfer, 96.3 KB of data
transferred 96.3 KB in * seconds (*/sec) (glob)
searching for changes
no changes found
Clone with background file closing enabled
$ hg --debug --config worker.backgroundclose=true --config worker.backgroundcloseminfilecount=1 clone --uncompressed -U http://localhost:$HGPORT clone-background | grep -v adding
using http://localhost:$HGPORT/
sending capabilities command
sending branchmap command
streaming all changes
sending stream_out command
1027 files to transfer, 96.3 KB of data
starting 4 threads for background file closing
transferred 96.3 KB in * seconds (*/sec) (glob)
query 1; heads
sending batch command
searching for changes
all remote heads known locally
no changes found
sending getbundle command
bundle2-input-bundle: with-transaction
bundle2-input-part: "listkeys" (params: 1 mandatory) supported
bundle2-input-part: "listkeys" (params: 1 mandatory) supported
bundle2-input-bundle: 1 parts total
checking for updated bookmarks
preparing listkeys for "phases"
sending listkeys command
received listkey for "phases": 58 bytes
Stream clone while repo is changing:
$ mkdir changing
$ cd changing
extension for delaying the server process so we reliably can modify the repo
while cloning
$ cat > delayer.py <<EOF
> import time
> from mercurial import extensions, scmutil
> def __call__(orig, self, path, *args, **kwargs):
> if path == 'data/f1.i':
> time.sleep(2)
> return orig(self, path, *args, **kwargs)
> extensions.wrapfunction(scmutil.vfs, '__call__', __call__)
> EOF
prepare repo with small and big file to cover both code paths in emitrevlogdata
$ hg init repo
$ touch repo/f1
$ $TESTDIR/seq.py 50000 > repo/f2
$ hg -R repo ci -Aqm "0"
$ hg -R repo serve -p $HGPORT1 -d --pid-file=hg.pid --config extensions.delayer=delayer.py
$ cat hg.pid >> $DAEMON_PIDS
clone while modifying the repo between stating file with write lock and
actually serving file content
$ hg clone -q --uncompressed -U http://localhost:$HGPORT1 clone &
$ sleep 1
$ echo >> repo/f1
$ echo >> repo/f2
$ hg -R repo ci -m "1"
$ wait
$ hg -R clone id
000000000000