core.fsyncmethod: add 'batch' mode for faster fsyncing of multiple objects #1134

This commit prepares for adding batch-fsync to the bulk-checkin infrastructure. The bulk-checkin infrastructure is currently used to batch up addition of large blobs to a packfile. When a blob is larger than big_file_threshold, we unconditionally add it to a pack. If bulk checkins are 'plugged', we allow multiple large blobs to be added to a single pack until we reach the packfile size limit; otherwise, we simply make a new packfile for each large blob. The 'unplug' call tells us when the series of blob additions is done so that we can finish the packfiles and make their objects available to subsequent operations. Stated another way, bulk-checkin allows callers to define a transaction that adds multiple objects to the object database, where the object database can optimize its internal operations within the transaction boundary. Batched fsync will fit into bulk-checkin by taking advantage of the plug/unplug functionality to determine the appropriate time to fsync and make newly-added objects available in the primary object database. * Rename 'state' variable to 'bulk_checkin_packfile', since we will later be adding 'bulk_fsync_objdir'. This also makes the variable easier to find in the debugger, since the name is more unique. * Rename finish_bulk_checkin to flush_bulk_checkin_packfile and call it unconditionally from unplug_bulk_checkin. Internally it will conditionally do a flush if there's any work to do. * Move the 'plugged' data member of 'bulk_checkin_state' into a separate static variable. Doing this avoids resetting the variable in finish_bulk_checkin when zeroing the 'bulk_checkin_state'. As-is, we seem to unintentionally disable the plugging functionality the first time a new packfile must be created due to packfile size limits. While disabling the plugging state only results in suboptimal behavior for the current code, it would be fatal for the bulk-fsync functionality later in this patch series. The net effect of these changes is to make a clear separation between the portion of the bulk-checkin infrastructure that is related to the packfile (nearly all of it at present) and the part that is related to other future optimizations of the ODB. Signed-off-by: Neeraj Singh <neerajsi@microsoft.com>

Make it clearer in the naming and documentation of the plug_bulk_checkin and unplug_bulk_checkin APIs that they can be thought of as a "transaction" to optimize operations on the object database. These transactions may be nested so that subsystems like the cache-tree writing code can optimize their operations without caring whether the top-level code has a transaction active. Add a flush_odb_transaction API that will be used in update-index to make objects visible even if a transaction is active. The flush call may also be useful in future cases if we hold a transaction active around calling hooks. Signed-off-by: Neeraj Singh <neerajsi@microsoft.com>

If we die while trying to fsync a loose object file, pass the actual filename we're trying to sync. This is likely to be more helpful for a user trying to diagnose the cause of the failure than the former 'loose object file' string. It also sidesteps any concerns about translating the die message differently for loose objects versus something else that has a real path. Signed-off-by: Neeraj Singh <neerajsi@microsoft.com>

When adding many objects to a repo with `core.fsync=loose-object`, the cost of fsync'ing each object file can become prohibitive. One major source of the cost of fsync is the implied flush of the hardware writeback cache within the disk drive. This commit introduces a new `core.fsyncMethod=batch` option that batches up hardware flushes. It hooks into the bulk-checkin odb-transaction functionality, takes advantage of tmp-objdir, and uses the writeout-only support code. When the new mode is enabled, we do the following for each new object: 1a. Create the object in a tmp-objdir. 2a. Issue a pagecache writeback request and wait for it to complete. At the end of the entire transaction when unplugging bulk checkin: 1b. Issue an fsync against a dummy file to flush the log and hardware writeback cache, which should by now have seen the tmp-objdir writes. 2b. Rename all of the tmp-objdir files to their final names. 3b. When updating the index and/or refs, we assume that Git will issue another fsync internal to that operation. This is not the default today, but the user now has the option of syncing the index and there is a separate patch series to implement syncing of refs. On a filesystem with a singular journal that is updated during name operations (e.g. create, link, rename, etc), such as NTFS, HFS+, or XFS we would expect the fsync to trigger a journal writeout so that this sequence is enough to ensure that the user's data is durable by the time the git command returns. This sequence also ensures that no object files appear in the main object store unless they are fsync-durable. Batch mode is only enabled if core.fsync includes loose-objects. If the legacy core.fsyncObjectFiles setting is enabled, but core.fsync does not include loose-objects, we will use file-by-file fsyncing. In step (1a) of the sequence, the tmp-objdir is created lazily to avoid work if no loose objects are ever added to the ODB. We use a tmp-objdir to maintain the invariant that no loose-objects are visible in the main ODB unless they are properly fsync-durable. This is important since future ODB operations that try to create an object with specific contents will silently drop the new data if an object with the target hash exists without checking that the loose-object contents match the hash. Only a full git-fsck would restore the ODB to a functional state where dataloss doesn't occur. In step (1b) of the sequence, we issue a fsync against a dummy file created specifically for the purpose. This method has a little higher cost than using one of the input object files, but makes adding new callers of this mechanism easier, since we don't need to figure out which object file is "last" or risk sharing violations by caching the fd of the last object file. _Performance numbers_: Linux - Hyper-V VM running Kernel 5.11 (Ubuntu 20.04) on a fast SSD. Mac - macOS 11.5.1 running on a Mac mini on a 1TB Apple SSD. Windows - Same host as Linux, a preview version of Windows 11. Adding 500 files to the repo with 'git add' Times reported in seconds. object file syncing | Linux | Mac | Windows --------------------|-------|-------|-------- disabled | 0.06 | 0.35 | 0.61 fsync | 1.88 | 11.18 | 2.47 batch | 0.15 | 0.41 | 1.53 Signed-off-by: Neeraj Singh <neerajsi@microsoft.com>

Take advantage of the odb transaction infrastructure around writing the cached tree to the object database. Signed-off-by: Neeraj Singh <neerajsi@microsoft.com>

The add_files_to_cache function is invoked internally by builtin/commit.c and builtin/checkout.c for their flags that stage modified files before doing the larger operation. These commands can benefit from batched fsyncing. Signed-off-by: Neeraj Singh <neerajsi@microsoft.com>

The update-index functionality is used internally by 'git stash push' to setup the internal stashed commit. This change enables odb-transactions for update-index infrastructure to speed up adding new objects to the object database by leveraging the batch fsync functionality. There is some risk with this change, since under batch fsync, the object files will be in a tmp-objdir until update-index is complete, so callers using the --stdin option will not see them until update-index is done. This risk is mitigated by flushing the ODB transaction prior to reporting any verbose output so that objects will be visible to callers that are synchronizing with update-index by snooping its output. Signed-off-by: Neeraj Singh <neerajsi@microsoft.com>

The unpack-objects functionality is used by fetch, push, and fast-import to turn the transfered data into object database entries when there are fewer objects than the 'unpacklimit' setting. By enabling an odb-transaction when unpacking objects, we can take advantage of batched fsyncs. Here are some performance numbers to justify batch mode for unpack-objects, collected on a WSL2 Ubuntu VM. Fsync Mode | Time for 90 objects (ms) ------------------------------------- Off | 170 On,fsync | 760 On,batch | 230 Note that the default unpackLimit is 100 objects, so there's a 3x benefit in the worst case. The non-batch mode fsync scales linearly with the number of objects, so there are significant benefits even with smaller numbers of objects. Signed-off-by: Neeraj Singh <neerajsi@microsoft.com>

Git for Windows has defaulted to core.fsyncObjectFiles=true since September 2017. We turn on syncing of loose object files with batch mode in upstream Git so that we can get broad coverage of the new code upstream. We don't actually do fsyncs in the most of the test suite, since GIT_TEST_FSYNC is set to 0. However, we do exercise all of the surrounding batch mode code since GIT_TEST_FSYNC merely makes the maybe_fsync wrapper always appear to succeed. Signed-off-by: Neeraj Singh <neerajsi@microsoft.com>

Several tests use awk to parse OIDs from the output of 'git ls-files --stage' and 'git ls-tree'. Introduce helpers to centralize these uses of awk. Update t5317-pack-objects-filter-objects.sh to use the new ls-files helper so that it has some usages to review. Other updates are left for the future. Signed-off-by: Neeraj Singh <neerajsi@microsoft.com>

Add test cases to exercise batch mode for: * 'git add' * 'git stash' * 'git update-index' * 'git unpack-objects' These tests ensure that the added data winds up in the object database. In this change we introduce a new test helper lib-unique-files.sh. The goal of this library is to create a tree of files that have different oids from any other files that may have been created in the current test repo. This helps us avoid missing validation of an object being added due to it already being in the repo. Signed-off-by: Neeraj Singh <neerajsi@microsoft.com>

Tests that affect the repo in stateful ways are easier to write if we can run setup steps outside of the measured portion of perf iteration. This change adds a "--setup 'setup-script'" parameter to test_perf. To make invocations easier to understand, I also moved the prerequisites to a new --prereq parameter. The setup facility will be used in the upcoming perf tests for batch mode, but it already helps in some existing tests, like t5302 and t7820. Signed-off-by: Neeraj Singh <neerajsi@microsoft.com>

Add basic performance tests for git commands that can add data to the object database. We cover: * git add * git stash * git update-index (via git stash) * git unpack-objects * git commit --all We cover all currently available fsync methods as well. Signed-off-by: Neeraj Singh <neerajsi@microsoft.com>

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

core.fsyncmethod: add 'batch' mode for faster fsyncing of multiple objects #1134

core.fsyncmethod: add 'batch' mode for faster fsyncing of multiple objects #1134

Commits on Mar 31, 2022