reftable [v2]: new ref storage format

reftable [v2]: new ref storage format

[Shawn Pearce]

This is an updated draft after discussion on list with Peff, Michael
Haggerty, and Dave Borowitz.

You can read a rendered version of this here:
https://googlers.googlesource.com/sop/jgit/+/reftable/Documentation/technical/reftable.md

Biggest changes from the first proposal are:

- reflog is now integrated into reftable
- block headers slightly different
- Peff's stack management idea is used
- Michael's compaction idea is used


## Overview

### Problem statement

Some repositories contain a lot of references (e.g.  android at 866k,
rails at 31k).  The existing packed-refs format takes up a lot of
space (e.g.  62M), and does not scale with additional references.
Lookup of a single reference requires linearly scanning the file.

Atomic pushes modifying multiple references require copying the
entire packed-refs file, which can be a considerable amount of data
moved (e.g. 62M in, 62M out) for even small transactions (2 refs
modified).

Repositories with many loose references occupy a large number of disk
blocks from the local file system, as each reference is its own file
storing 41 bytes (and another file for the corresponding reflog).
This negatively affects the number of inodes available when a large
number of repositories are stored on the same filesystem.  Readers can
be penalized due to the larger number of syscalls required to traverse
and read the `$GIT_DIR/refs` directory.

### Objectives

- Near constant time lookup for any single reference, even when the
  repository is cold and not in process or kernel cache.
- Efficient lookup of an entire namespace, such as `refs/tags/`.
- Occupy less disk space for large repositories.
- Support atomic push `O(size_of_update)` operations.
- Combine reflog storage with ref storage.

### Description

A reftable file is a portable binary file format customized for
reference storage. References are sorted, enabling linear scans,
binary search lookup, and range scans.

Storage in the file is organized into blocks.  Prefix compression
is used within a single block to reduce disk space.  Block size is
tunable by the writer.

### Performance

Space used, packed-refs vs. reftable:

repository | packed-refs | reftable | % original | avg ref
-----------|------------:|---------:|-----------:|---------:
android    |      62.2 M |   27.7 M |     44.4%  | 33 bytes
rails      |       1.8 M |  896.2 K |     47.6%  | 29 bytes
git        |      78.7 K |   27.9 K |     40.0%  | 43 bytes
git (heads)|       332 b |    222 b |     66.9%  | 37 bytes

Scan (read 866k refs) and lookup (single ref from 866k refs):

format      | scan    | lookup
------------|--------:|---------------:
packed-refs |  380 ms | 375420.0 usec
reftable    |  125 ms |     42.3 usec

Space used for 149,932 log entries on 43,061 refs,
reflog vs. reftable:

format        | size  | avg log
--------------|------:|-----------:
$GIT_DIR/logs | 173 M | 1209 bytes
reftable      |   4 M |   30 bytes

## Details

### Peeling

References in a reftable are always peeled.

### Reference name encoding

Reference names should be encoded with UTF-8.

### Ordering

Blocks are lexicographically ordered by their first reference.

### Directory/file conflicts

The reftable format accepts both `refs/heads/foo` and
`refs/heads/foo/bar` as distinct references.

This property is useful for retaining log records in reftable, but may
confuse versions of Git using `$GIT_DIR/refs` directory tree to
maintain references.  Users of reftable may choose to continue to
reject `foo` and `foo/bar` type conflicts to prevent problems for
peers.

## File format

### Structure

A reftable file has the following basic structure:

    first_block {
      header
      first_ref_block
    }
    ref_blocks*
    ref_index?
    log_blocks*
    log_index?
    footer

### Block size

The `block_size` is arbitrarily determined by the writer, and does not
have to be a power of 2.  The block size must be larger than the
longest reference name or log entry used in the repository, as
references cannot span blocks.

Powers of two that are friendly to the virtual memory system or
filesystem (such as 4k or 8k) are recommended.  Larger sizes
(64k) can yield better compression.

The largest block size is `16777215` bytes (15.99 MiB).

### Header

A 8-byte header appears at the beginning of the file:

    '\1REF'
    uint8( version_number = 1 )
    uint24( block_size )

The `block_size` and all other uint fields are in network byte order.

### First ref block

The first ref block shares the same block as the file header, and is 8
bytes smaller than all other blocks in the file.  The first block
immediately begins after the file header, at offset 8.

### Ref block format

A ref block is written as:

    'r'
    uint24 ( block_len )
    ref_record*
    uint32( restart_offset )*
    uint16( number_of_restarts )
    padding?

Blocks begin with `block_type = 'r'` and a 3-byte `block_len` which
encodes the number of bytes in the block up to, but not including the
optional `padding`.  This is almost always shorter than the file's
`block_size`.  In the first ref block, `block_len` includes 8 bytes
for the file header.

The 4-byte block header is followed by a variable number of
`ref_record`, describing reference names and values.  The format
is described below.

A variable number of 4-byte `restart_offset` values follows the
records.  Offsets are relative to the start of the block (0 in first
block to include file header) and refer to the first byte of any
`ref_record` whose name has not been prefixed compressed.  Readers can
start linear scans from any of these records.

The 2-byte `number_of_restarts + 1` stores the number of entries in
the `restart_offset` list.

Readers can use the restart count to binary search between restarts
before starting a linear scan.  The `number_of_restarts` field must be
the last 2 bytes of the block as specified by `block_len` from the
block header.

The end of the record may be filled with `padding` NUL bytes to fill
out the block to the common `block_size` as specified in the file
header.  Padding may be necessary to ensure the following block starts
at a block alignment, and does not spill into the tail of this block.
Padding may be omitted if the block is the last block of the file, or
there is no index block.  This allows reftable to efficiently scale
down to a small number of refs.

#### ref record

A `ref_record` describes a single reference, storing both the name and
its value(s). Records are formatted as:

    varint( prefix_length )
    varint( (suffix_length << 2) | value_type )
    suffix
    value?

The `prefix_length` field specifies how many leading bytes of the
prior reference record's name should be copied to obtain this
reference's name.  This must be 0 for the first reference in any
block, and also must be 0 for any `ref_record` whose offset is listed
in the `restart_offset` table at the end of the block.

Recovering a reference name from any `ref_record` is a simple concat:

    this_name = prior_name[0..prefix_length] + suffix

The second varint carries both `suffix_length` and `value_type`.  The
`suffix_length` value provides the number of bytes to copy from
`suffix` to complete the reference name.

The `value` immediately follows.  Its format is determined by
`value_type`, a 2 bit code, one of the following:

- `0x0`: deletion; no value data (see transactions, below)
- `0x1`: one 20-byte object id; value of the ref
- `0x2`: two 20-byte object ids; value of the ref, peeled target
- `0x3`: symbolic reference: `varint( target_len ) target`

Symbolic references use a varint length followed by a variable number
of bytes to encode the complete reference target.  No compression is
applied to the target name.

### Ref index

The ref index stores the name of the last reference from every ref
block in the file, enabling constant O(1) disk seeks for all lookups.
Any reference can be found by binary searching the index, identifying
the containing block, and searching within that block.

If present, the ref index block appears after the last ref block.  The
prior ref block should be padded to ensure the ref index starts on a
block alignment.

An index block should only be written if there are at least 4 blocks
in the file, as cold reads using the index requires 2 disk reads, and
binary searching <= 4 blocks also requires <= 2 reads.  Omitting the
index block from smaller files saves space.

Index block format:

    uint32( (0x80 << 24) | block_len )
    index_record*
    uint32( restart_offset )*
    uint16( number_of_restarts )

The index block header starts with the high bit set.  This identifies
the block as an index block, and not as a ref block, log block or file
footer.  The `block_len` field in an index block is 30-bits network
byte order, and allowed to occupy the space normally used by the block
type in other blocks.  This supports indexes significantly larger than
the file's `block_size`.

The `restart_offset` and `number_of_restarts` fields are identical in
format, meaning and usage as in ref blocks.

To reduce the number of reads required for random access in very large
files, the index block may be larger than the other blocks.  However,
readers must hold the entire index in memory to benefit from this, so
its a time-space tradeoff in both file size, and reader memory.
Increasing the block size in the writer decreases the index size.

Unlike ref blocks, the index block is not padded.

#### index record

An index record describes the last entry in another block.
Index records are written as:

    varint( prefix_length )
    varint( (suffix_length << 2) )
    suffix
    varint( block_offset )

Index records use prefix compression exactly like `ref_record`.  The
`suffix_length` is shifted 2 bits without a `type` to simplify unified
reader/writer code for both block types.

Index records store `block_offset` after the suffix, specifying the
offset in bytes (from the start of the file) of the block that ends
with this reference.

#### Reading the index

Readers loading the ref index must first read the footer (below) to
obtain `ref_index_offset`. If not present, the offset will be 0.

### Log block format

A log block is written as:

    'g'
    uint24( block_len )
    zlib_deflate {
      log_record*
      int32( restart_offset )*
      int16( number_of_restarts )
    }

Log blocks look similar to ref blocks, except `block_type = 'g'`.

The 4-block header is followed by the deflated block contents using
zlib deflate.  The `block_len` in the header is the inflated size
(including 4-byte block header), and should be used by readers to
preallocate the inflation output buffer.  Offsets within the block
(e.g.  `restart_offset`) still include the 4-byte header.  Readers may
prefer prefixing the inflation output buffer with the 4-byte header.

Within the deflate container, a variable number of `log_record`,
describing reference changes.  The log record format is described
below.  See ref block format (above) for a description of
`restart_offset` and `number_of_restarts`.

Unlike ref blocks, log blocks are written at any alignment, without
padding.  The first log block immediately follows the end of the last
ref block, or the ref index.  In very small files the log block may
appear in the first block.

Readers must keep track of the bytes consumed by the inflater to know
where the next log block begins.

#### log record

Log record keys are structured as:

    ref_name '\0' reverse_int32( time_sec )

where `time_sec` is the update time in seconds since the epoch.  The
`reverse_int32` function inverses the value so lexographical ordering
the network byte order time sorts more recent records first:

    reverse_int(int32 t) {
      return 0xffffffff - t;
    }

Log records have a similar starting structure to ref and index
records, utilizing the same prefix compression scheme applied to the
key described above.  Like in an index record, a log record uses
`value_type = 0x0`:

    varint( prefix_length )
    varint( (suffix_length << 2) | 0x0 )
    suffix

    old_id
    new_id
    sint16( tz_offset )
    varint( name_length  )   name
    varint( email_length )   email
    varint( message_length ) message

The value data following the key suffix is complex:

- two 20-byte SHA-1s (old id, new id)
- 2-byte timezone offset (signed)
- varint string of committer's name
- varint string of committer's email
- varint string of message

`tz_offset` is the absolute number of minutes from GMT the committer
was at the time of the update.  For example `GMT-0800` is encoded in
reftable as `int16(-480)` and `GMT+0230` is `int16(150)`.

The `message_length` may be 0, in which case there was no message
supplied for the update.

#### Reading the log

Readers accessing the log must first read the footer (below) to
determine the `log_offset`.  The first block of the log begins at
`log_offset` bytes since the start of the file.

In very small reftable files `log_offset`, may not be block aligned.

### Log index

The log index stores the log key (`refname \0 reverse_int32(time_sec)`)
for the last log record of every log block in the file, supporting
bounded-time lookup.

A log index block must be written if 2 or more log blocks are written
to the file.  If present, the log index appears after the first log
block.  There is no padding used to align the log index to block
alignment.

Log index format is identical to ref index, except the keys are 5
bytes longer to include `'\0'` and the 4-byte `reverse_int32(time)`.
Records use `block_offset` to refer to the start of a log block.

#### Reading the index

Readers loading the log index must first read the footer (below) to
obtain `log_index_offset`. If not present, the offset will be 0.

### Footer

After the last block of the file, a file footer is written.  It begins
like the file header, but is extended with additional data.

A 36-byte footer appears at the end:

    '\1REF'
    uint8( version_number = 1 )
    uint24( block_size )

    uint64( ref_index_offset )
    uint64( log_offset )
    uint64( log_index_offset )
    uint32( CRC-32 of prior )

If a section is missing (e.g. ref index) the corresponding offset
field (e.g. `ref_index_offset`) will be 0.

#### Reading the footer

Readers must seek to `file_length - 36` to access the footer.  A
trusted external source (such as `stat(2)`) is necessary to obtain
`file_length`.  When reading the footer, readers must verify:

- 4-byte magic is correct
- 1-byte version number is recognized
- 4-byte CRC-32 matches the other 32 bytes (including magic, and version)

Once verified, the other fields of the footer can be accessed.

### Varint encoding

Varint encoding is identical to the ofs-delta encoding method used
within pack files.

Decoder works such as:

    val = buf[ptr] & 0x7f
    while (buf[ptr] & 0x80) {
      ptr++
      val++
      val = val << 7
      val = val | (buf[ptr] & 0x7f)
    }

### Binary search

Binary search within a block is supported by the `restart_offset`
fields at the end of the block.  Readers can binary search through the
restart table to locate between which two restart points the sought
reference or key should appear.

Each record identified by a `restart_offset` stores the complete key
in the `suffix` field of the record, making the compare operation
during binary search straightforward.

Once a restart point lexicographically before the sought reference has
been identified, readers can linearly scan through the following
record entries to locate the sought record, terminating if the current
record sorts after (and therefore the sought key is not present).

#### Restart point selection

Writers determine the restart points at file creation.  The process is
arbitrary, but every 16 or 64 records is recommended.  Every 16 may
be more suitable for smaller block sizes (4k or 8k), every 64 for
larger block sizes (64k).

More frequent restart points reduces prefix compression and increases
space consumed by the restart table, both of which increase file size.

Less frequent restart points makes prefix compression more effective,
decreasing overall file size, with increased penalities for readers
walking through more records after the binary search step.

A maxium of `65536` restart points is supported in any type of block.
The `restart_count` field stores restart points `-1`, with a maxium
value of `65535`.

## Considerations

### Lightweight refs dominate

The reftable format assumes the vast majority of references are single
SHA-1 valued with common prefixes, such as Gerrit Code Review's
`refs/changes/` namespace, GitHub's `refs/pulls/` namespace, or many
lightweight tags in the `refs/tags/` namespace.

Annotated tags storing the peeled object cost only an additional 20
bytes per reference.

### Low overhead

A reftable with very few references (e.g.  git.git with 5 heads) uses
only 222 bytes for reftable vs. 332 bytes for packed-refs.  This
supports reftable scaling down, to be used for transaction logs
(below).

### Block size

For a Gerrit Code Review type repository with many change refs, larger
block sizes (64 KiB) and less frequent restart points (every 64) yield
better compression due to more references within the block able to
compress against the prior reference.

Larger block sizes reduces the index size, as the reftable will
require fewer blocks to store the same number of references.

### Minimal disk seeks

Assuming the index block has been loaded into memory, binary searching
for any single reference requires exactly 1 disk seek to load the
containing block.

### Logs are infrequently read

Logs are infrequently accessed, but can be large.  Deflating log
blocks saves disk space, with some increased penalty at read time.

Logs are stored in an isolated section from refs, reducing the burden
on reference readers that want to ignore logs.

### Logs are read backwards

Logs are frequently accessed backwards (most recent N records for master),
so log records are grouped by reference, and sorted descending by time.

## Repository format

When reftable is stored in a file-backed Git repository, the stack is
represented as a series of reftable files in the dedicated
`$GIT_DIR/reftable/` directory:

    stack
    22b9ac56abe0ddc5f23d651fe4ef46343f338d20.ref
    82f128e165c8afae9894aacf0219bc53a580f2ad.ref

where `.ref` files are named by the SHA-1 hash of the contents of the
reftable.

The `stack` file lists the current files, one per line, in order, from
oldest (base) to newest (most recent):

    $ cat stack
    82f128e165c8afae9894aacf0219bc53a580f2ad.ref
    22b9ac56abe0ddc5f23d651fe4ef46343f338d20.ref

Readers must read `stack` to determine which files are relevant right
now, and search through the stack in reverse order (last reftable is
examined first).

Files not listed in `stack` may be new (and about to be added to the
stack by the active writer), or ancient and ready to be pruned.

### Update transactions

Although reftables are immutable, mutations are supported by writing a
new reftable and atomically appending it to the stack:

1. Atomically create `stack.lock`
2. Copy current stack to `stack.lock`.
3. Prepare new reftable in temp file `.refXXXXXXX`.
   Include log entries.
4. Rename (3) to `${sha1}.ref`.
5. Append `${sha1}.ref` to `stack.lock`
6. Atomically rename `stack.lock` to `stack`.

Because a single `stack.lock` file is used to manage locking, the
repository is single-threaded for writers.  Writers may have to
busy-spin (with some small backoff) around creating `stack.lock`,
for up to an acceptable wait period, aborting if the repository is too
busy to mutate.  Application servers wrapped around repositories (e.g.
Gerrit Code Review) can layer their own in memory thread lock/wait
queue to improve fairness.

### Reference deletions

Deletion of any reference can be explicitly stored by setting the
`type` to `0x0` and omitting the `value` field of the `ref_record`.
This entry shadows the reference in earlier files in the stack.

### Compaction

A partial stack of reftables can be compacted by merging references
using a straightforward merge join across reftables, selecting the
most recent value for output, and omitting deleted references that do
not appear in remaining, lower reftables.

For sake of illustration, assume the stack currently consists of
reftable files (from oldest to newest): A, B, C, and D. The compactor
is going to compact B and C, leaving A and D alone.

1.  Obtain lock `stack.lock` and read the `stack` file.
2.  Obtain locks `B.lock` and `C.lock`.
    Ownership of these locks prevents other processes from trying
to compact these files.
3.  Release `stack.lock`.
4.  Compact `B` and `C` in temp file `.refXXXXXXX`.
5.  Reacquire lock `stack.lock`.
6.  Verify that `B` and `C` are still in the stack, in that order. This
    should always be the case, assuming that other processes are adhering
    to the locking protocol.
7.  Rename `.refXXXXXXX` to `X`.
8.  Write the new stack to `stack.lock`, replacing `B` and `C` with `X`.
9.  Atomically rename `stack.lock` to `stack`.
10. Delete `B` and `C`, perhaps after a short sleep to avoid forcing
    readers to backtrack.

This strategy permits compactions to proceed independently of updates.

## Alternatives considered

### bzip packed-refs

`bzip2` can significantly shrink a large packed-refs file (e.g. 62
MiB compresses to 23 MiB, 37%).  However the bzip format does not support
random access to a single reference. Readers must inflate and discard
while performing a linear scan.

Breaking packed-refs into chunks (individually compressing each chunk)
would reduce the amount of data a reader must inflate, but still
leaves the problem of indexing chunks to support readers efficiently
locating the correct chunk.

Given the compression ratios achieved by reftable's simple encoding
(e.g.  44%), without using a standard compression algorithm, it does
not seem necessary to add the complexity of bzip/gzip/zlib.

### JGit Ketch RefTree

[JGit Ketch][ketch] proposed [RefTree][reftree], an encoding of
references inside Git tree objects stored as part of the repository's
object database.

The RefTree format adds additional load on the object database storage
layer (more loose objects, more objects in packs), and relies heavily
on the packer's delta compression to save space.  Namespaces which are
flat (e.g.  thousands of tags in refs/tags) initially create very
large loose objects, and so RefTree does not address the problem of
copying many references to modify a handful.

Flat namespaces are not efficiently searchable in RefTree, as tree
objects in canonical formatting cannot be binary searched. This fails
the need to handle a large number of references in a single namespace,
such as GitHub's `refs/pulls`, or a project with many tags.

[ketch]: https://dev.eclipse.org/mhonarc/lists/jgit-dev/msg03073.html
[reftree]: https://public-inbox.org/git/CAJo=hJvnAPNAdDcAAwAvU9C4RVeQdoS3Ev9WTguHx4fD0V_nOg@mail.gmail.com/

### LMDB

David Turner proposed [using LMDB][dt-lmdb], as LMDB is lightweight
(64k of runtime code) and GPL-compatible license.

A downside of LMDB is its reliance on a single C implementation.  This
makes embedding inside JGit (a popular reimplemenation of Git)
difficult, and hoisting onto virtual storage (for JGit DFS) virtually
impossible.

A common format that can be supported by all major Git implementations
(git-core, JGit, libgit2) is strongly preferred.

[dt-lmdb]: https://public-inbox.org/git/1455772670-21142-26-git-send-email-dturner@twopensource.com/

## Future

### Longer hashes

Version will bump (e.g.  2) to indicate `value` uses a different
object id length other than 20.  The length could be stored in an
expanded file header, or hardcoded as part of the version.

Re: reftable [v2]: new ref storage format

[Stefan Beller]

On Mon, Jul 17, 2017 at 8:01 AM, Shawn Pearce <spearce@spearce.org> wrote:

> A ref block is written as:
>
>     'r'
>     uint24 ( block_len )
>     ref_record*
>     uint32( restart_offset )*
>     uint16( number_of_restarts )
>     padding?
>
...
>
> The 2-byte `number_of_restarts + 1` stores the number of entries in
> the `restart_offset` list.

This means uint16( number_of_restarts ) cannot be 0, but has to be 1
to indicate no restarts?

Why do we need to be non-NUL here, but the padding is all NUL?

When starting to write a block, we need to know exactly how large
the ref_records* and restart offsets need to be to put the
number_of_restarts at the position as promised via block_len.
This sounds complicated unless I missed the obvious.

Going by this, would it rather make sense to omit the block_len
and then scan backwards from *block_size-1 to find the first non-NUL
and that will be the number_of_restarts?

Or we could allow additional padding between ref_record and
restart_offsets, such that the writer implementation has wiggle room
for the restarting logic.

>
> #### log record
>
> Log record keys are structured as:
>
>     ref_name '\0' reverse_int32( time_sec )
>
> where `time_sec` is the update time in seconds since the epoch.

The epoch ends 2038, which is in 21 years. (Did you just volunteer
to fixup the time issues in 20 years?)
If possible I'd prefer a date to be encoded with more range available.

>  The
> `reverse_int32` function inverses the value so lexographical ordering
> the network byte order time sorts more recent records first:
>
>     reverse_int(int32 t) {
>       return 0xffffffff - t;
>     }
>
> Log records have a similar starting structure to ref and index
> records, utilizing the same prefix compression scheme applied to the
> key described above.

The ref names itself are compressed, would we also want to compress
the timing information? The time_sec could be a varint indicating a delta
to the previous change of a ref, or fixed to the epoch if it is the first change
of that ref.

Just to be clear, the ordering here would be

  refs/heads/maint <number>
  refs/heads/maint <smaller number>
  ...
  refs/heads/master <number>
  refs/heads/master <smaller number>

such that refs that have more than one entry in its reflog in a given
refstable file, would have perfect prefix compression for the ref?

> ### Update transactions
>
> Although reftables are immutable, mutations are supported by writing a
> new reftable and atomically appending it to the stack:
>
> 1. Atomically create `stack.lock`
> 2. Copy current stack to `stack.lock`.
> 3. Prepare new reftable in temp file `.refXXXXXXX`.
>    Include log entries.
> 4. Rename (3) to `${sha1}.ref`.
> 5. Append `${sha1}.ref` to `stack.lock`
> 6. Atomically rename `stack.lock` to `stack`.

In case 3.+4. becomes time consuming, it can be prepared outside
the lock, such that inside the lock we'd only need to check
for contention of refs? For example if I'd want to update one ref and
another write wants to update another refs, we'd both be preparing
the a new refstable containing one ref and log, then one of us obtains
the lock and writes. The second writer would then need to inspect
the delta of the stack and see if any ref that they want to change
was touched.

> ### Compaction
>
> A partial stack of reftables can be compacted by merging references
> using a straightforward merge join across reftables, selecting the
> most recent value for output, and omitting deleted references that do
> not appear in remaining, lower reftables.
>
> For sake of illustration, assume the stack currently consists of
> reftable files (from oldest to newest): A, B, C, and D. The compactor
> is going to compact B and C, leaving A and D alone.
>
> 1.  Obtain lock `stack.lock` and read the `stack` file.
> 2.  Obtain locks `B.lock` and `C.lock`.
>     Ownership of these locks prevents other processes from trying
> to compact these files.
> 3.  Release `stack.lock`.
> 4.  Compact `B` and `C` in temp file `.refXXXXXXX`.
> 5.  Reacquire lock `stack.lock`.
> 6.  Verify that `B` and `C` are still in the stack, in that order. This
>     should always be the case, assuming that other processes are adhering
>     to the locking protocol.
> 7.  Rename `.refXXXXXXX` to `X`.
> 8.  Write the new stack to `stack.lock`, replacing `B` and `C` with `X`.
> 9.  Atomically rename `stack.lock` to `stack`.
> 10. Delete `B` and `C`, perhaps after a short sleep to avoid forcing
>     readers to backtrack.
>
> This strategy permits compactions to proceed independently of updates.

10. could be deferred to gc as well. auto gc would need to learn about
the number
of loose refstables in that case.

Thanks,
Stefan

Re: reftable [v2]: new ref storage format

[Shawn Pearce]

On Mon, Jul 17, 2017 at 11:53 AM, Stefan Beller <sbeller@google.com> wrote:
> On Mon, Jul 17, 2017 at 8:01 AM, Shawn Pearce <spearce@spearce.org> wrote:
>
>> A ref block is written as:
>>
>>     'r'
>>     uint24 ( block_len )
>>     ref_record*
>>     uint32( restart_offset )*
>>     uint16( number_of_restarts )
>>     padding?
>>
> ...
>>
>> The 2-byte `number_of_restarts + 1` stores the number of entries in
>> the `restart_offset` list.
>
> This means uint16( number_of_restarts ) cannot be 0, but has to be 1
> to indicate no restarts?

A block must have at least one restart in it, the first ref_record
must be a restart. So number_of_restarts in the tail of the block can
be 0, which implies 1 restart (number_of_restarts + 1), and the first
restart is required at the first ref_record. :)

> When starting to write a block, we need to know exactly how large
> the ref_records* and restart offsets need to be to put the
> number_of_restarts at the position as promised via block_len.
> This sounds complicated unless I missed the obvious.

Correct. The writer needs to compute the block size before it writes
the block. It does so by buffering the block contents until its
approximately full, then fixes block_len, and flushes the block.

> Going by this, would it rather make sense to omit the block_len
> and then scan backwards from *block_size-1 to find the first non-NUL
> and that will be the number_of_restarts?

Not quite. On small reftable files the "physical" block may be shared
with a log block ('g'). We need to be able to reliably find the of the
ref block ('r'), without padding between the two blocks.

> Or we could allow additional padding between ref_record and
> restart_offsets, such that the writer implementation has wiggle room
> for the restarting logic.

I had that in an older format description, and removed it. Placing the
padding at the end of the block was simpler for the reader and writer
implementation to handle.


>> Log record keys are structured as:
>>
>>     ref_name '\0' reverse_int32( time_sec )
>>
>> where `time_sec` is the update time in seconds since the epoch.
>
> The epoch ends 2038, which is in 21 years. (Did you just volunteer
> to fixup the time issues in 20 years?)
> If possible I'd prefer a date to be encoded with more range available.

Good point. However, I think in 20 years we'll see 2 more hash
functions for Git, and we can bump reftable to v2 and expand the field
to 8 bytes.

> The ref names itself are compressed, would we also want to compress
> the timing information?

The time field is also prefix compressed as part of the ref name's
prefix compression. So there is no need to move to the complexity of a
varint or anything else.


>> ### Update transactions
>>
>> Although reftables are immutable, mutations are supported by writing a
>> new reftable and atomically appending it to the stack:
>>
>> 1. Atomically create `stack.lock`
>> 2. Copy current stack to `stack.lock`.
>> 3. Prepare new reftable in temp file `.refXXXXXXX`.
>>    Include log entries.
>> 4. Rename (3) to `${sha1}.ref`.
>> 5. Append `${sha1}.ref` to `stack.lock`
>> 6. Atomically rename `stack.lock` to `stack`.
>
> In case 3.+4. becomes time consuming, it can be prepared outside
> the lock, such that inside the lock we'd only need to check
> for contention of refs? For example if I'd want to update one ref and
> another write wants to update another refs, we'd both be preparing
> the a new refstable containing one ref and log, then one of us obtains
> the lock and writes. The second writer would then need to inspect
> the delta of the stack and see if any ref that they want to change
> was touched.

Excellent point, it reduces the contention window for non-conflicting
writes. I will update this section with your input, thank you Stefan.

Re: reftable [v2]: new ref storage format

[Stefan Beller]

On Mon, Jul 17, 2017 at 12:04 PM, Shawn Pearce <spearce@spearce.org> wrote:

> A block must have at least one restart in it, the first ref_record
> must be a restart. So number_of_restarts in the tail of the block can
> be 0, which implies 1 restart (number_of_restarts + 1), and the first
> restart is required at the first ref_record. :)

Hah! I assumed the first entry to not be recorded because it is always
a restart by definition of the file format, so it could be omitted in the
restart_offset list, but that would complicate the implementation, such
that including it makes sense.

>
>> When starting to write a block, we need to know exactly how large
>> the ref_records* and restart offsets need to be to put the
>> number_of_restarts at the position as promised via block_len.
>> This sounds complicated unless I missed the obvious.
>
> Correct. The writer needs to compute the block size before it writes
> the block. It does so by buffering the block contents until its
> approximately full, then fixes block_len, and flushes the block.

So that is another trade off for determining the block size. "How much
can I buffer?"

>> Going by this, would it rather make sense to omit the block_len
>> and then scan backwards from *block_size-1 to find the first non-NUL
>> and that will be the number_of_restarts?
>
> Not quite. On small reftable files the "physical" block may be shared
> with a log block ('g'). We need to be able to reliably find the of the
> ref block ('r'), without padding between the two blocks.

I'd need to reread the proposal to understand this bit as I assumed that
each block starts at a multiple of block_size. However we could choose
block_size such that there is no padding between 'r' and 'g'. Ok,
makes sense.

> The time field is also prefix compressed as part of the ref name's
> prefix compression. So there is no need to move to the complexity of a
> varint or anything else.

I agree, that is why you explicitly said that the key is
    ref_name '\0' reverse_int32( time_sec )

Note (as found out in discussion with jrnieder@): The size of the integer
is determined by the suffix length encoding and the preceding '\0',
such that the file format allows arbitrary integer size. So instead of
pretending we can only do 32 bit here, just say 'uint' ?

Re: reftable [v2]: new ref storage format

[Junio C Hamano]

Shawn Pearce <spearce@spearce.org> writes:

> This is an updated draft after discussion on list with Peff, Michael
> Haggerty, and Dave Borowitz.
>
> You can read a rendered version of this here:
> https://googlers.googlesource.com/sop/jgit/+/reftable/Documentation/technical/reftable.md
>
> Biggest changes from the first proposal are:
>
> - reflog is now integrated into reftable
> - block headers slightly different
> - Peff's stack management idea is used
> - Michael's compaction idea is used

Just a few comments.

> A variable number of 4-byte `restart_offset` values follows the
> records.  Offsets are relative to the start of the block (0 in first
> block to include file header) and refer to the first byte of any
> `ref_record` whose name has not been prefixed compressed.  Readers can
> start linear scans from any of these records.

It is unclear what "0 in first block to include file header" wants
to say.  Do I write "8" if I want to express the offset of the first
record in the first block, or do I write "0"?

> The 2-byte `number_of_restarts + 1` stores the number of entries in
> the `restart_offset` list.

It is unclear whose responsibility it is to add this "1".  Does this
mean a reader thinks there is one entry in the restart table when it
sees "0" in the number_of_restarts field (hence you can have max
65536 entries in total)?

> Readers can use the restart count to binary search between restarts
> before starting a linear scan.  The `number_of_restarts` field must be
> the last 2 bytes of the block as specified by `block_len` from the
> block header.

Does the new term "restart count" mean the same thing as
number_of_restarts?

> ### Log block format
>
> A log block is written as:
>
>     'g'
>     uint24( block_len )
>     zlib_deflate {
>       log_record*
>       int32( restart_offset )*
>       int16( number_of_restarts )
>     }
>
> Log blocks look similar to ref blocks, except `block_type = 'g'`.

Is there a general recommended strategy for writers to choose how
many entries to include in a single physical block?  I understand
that the deflated whole must fit in the physical block whose length
is defined in the footer of the whole file, and in general you would
not know how small the data deflates down to before compressing,
right?

> Log record keys are structured as:
>
>     ref_name '\0' reverse_int32( time_sec )
>
> where `time_sec` is the update time in seconds since the epoch.  The
> `reverse_int32` function inverses the value so lexographical ordering
> the network byte order time sorts more recent records first:
>
>     reverse_int(int32 t) {
>       return 0xffffffff - t;
>     }

Is 2038 an issue, or by that time we'd all be retired together with
this file format and it won't be our problem?

As the file format uses delta compression with restarts, a reader
needs to sequencially scan some bounded number of entries to get the
contents of a specific entry anyway, so I am wondering if it is
worth storing a longer timestamp in varint() for an restart entry
and express the timestamp on delta entries as difference to the
previous entry.

> ### Log index
>
> The log index stores the log key (`refname \0 reverse_int32(time_sec)`)
> for the last log record of every log block in the file, supporting
> bounded-time lookup.

This assumes that timestamps never wildly rewind in the reflog,
doesn't it?  Is that a sensible assumption?

Or does "the last log record" in the above mean "the log record with
largest timestamp?  ... ah, not that is still not sufficient.  You'd
still need to assume that timestamps on entries in one log block must
be all older than the ones on entries in later log blocks.  Hmph...

Also it is not clear to me how reflogs for two refs would be
intermixed in the log blocks, and what log keys for the entries must
be recorded in the log index, to facilitate efficient lookup.  Is it
assumed that a consecutive sequence of log blocks record reflogs for
the same ref, before the sequence of log blocks switch to record
reflogs for another ref, or something?

> A log index block must be written if 2 or more log blocks are written
> to the file.  If present, the log index appears after the first log
> block.  There is no padding used to align the log index to block
> alignment.
>
> Log index format is identical to ref index, except the keys are 5
> bytes longer to include `'\0'` and the 4-byte `reverse_int32(time)`.
> Records use `block_offset` to refer to the start of a log block.

I am assuming that we do not care about being able to quickly
determine master@{24028}; I would imagine that it wouldn't be too
hard to add an index to help such query, but I offhand would not
know the details until I figure out how the format handles reflog
entries for multiple refs first.

Re: reftable [v2]: new ref storage format

[Shawn Pearce]

On Mon, Jul 17, 2017 at 12:51 PM, Junio C Hamano <gitster@pobox.com> wrote:
> Shawn Pearce <spearce@spearce.org> writes:
>> You can read a rendered version of this here:
>> https://googlers.googlesource.com/sop/jgit/+/reftable/Documentation/technical/reftable.md
>
> Just a few comments.
>
>> A variable number of 4-byte `restart_offset` values follows the
>> records.  Offsets are relative to the start of the block (0 in first
>> block to include file header) and refer to the first byte of any
>> `ref_record` whose name has not been prefixed compressed.  Readers can
>> start linear scans from any of these records.
>
> It is unclear what "0 in first block to include file header" wants
> to say.  Do I write "8" if I want to express the offset of the first
> record in the first block, or do I write "0"?

"8". I've updated the text to try and clarify this better.


>> The 2-byte `number_of_restarts + 1` stores the number of entries in
>> the `restart_offset` list.
>
> It is unclear whose responsibility it is to add this "1".  Does this
> mean a reader thinks there is one entry in the restart table when it
> sees "0" in the number_of_restarts field (hence you can have max
> 65536 entries in total)?

Correct, I've reworded this section to clarify the reader must add +1
to reach the potential max of 65536 entries in total.

>> Readers can use the restart count to binary search between restarts
>> before starting a linear scan.  The `number_of_restarts` field must be
>> the last 2 bytes of the block as specified by `block_len` from the
>> block header.
>
> Does the new term "restart count" mean the same thing as
> number_of_restarts?

Not quite (because of the +1 issue), I've fixed the document to
introduce and define restart_count.


>> ### Log block format
>>
>> A log block is written as:
>>
>>     'g'
>>     uint24( block_len )
>>     zlib_deflate {
>>       log_record*
>>       int32( restart_offset )*
>>       int16( number_of_restarts )
>>     }
>>
>> Log blocks look similar to ref blocks, except `block_type = 'g'`.
>
> Is there a general recommended strategy for writers to choose how
> many entries to include in a single physical block?  I understand
> that the deflated whole must fit in the physical block whose length
> is defined in the footer of the whole file, and in general you would
> not know how small the data deflates down to before compressing,
> right?

No, this is an  incorrect understanding. The deflated log blocks do
not match the physical block length of the file. They are variable
length, matching whatever the deflater output, with no inter-block
padding.

Writers should allocate a "reasonable buffer" (my prototype has it
default to 2x the ref block length), pack log records into that, then
deflate that when its at capacity.


>> Log record keys are structured as:
>>
>>     ref_name '\0' reverse_int32( time_sec )
>>
>> where `time_sec` is the update time in seconds since the epoch.  The
>> `reverse_int32` function inverses the value so lexographical ordering
>> the network byte order time sorts more recent records first:
>>
>>     reverse_int(int32 t) {
>>       return 0xffffffff - t;
>>     }
>
> Is 2038 an issue, or by that time we'd all be retired together with
> this file format and it won't be our problem?

Based on discussion with Michael Haggerty, this is now an 8 byte field
storing microseconds since the epoch. We should be good through year
9999.


>> ### Log index
>>
>> The log index stores the log key (`refname \0 reverse_int32(time_sec)`)
>> for the last log record of every log block in the file, supporting
>> bounded-time lookup.
>
> This assumes that timestamps never wildly rewind in the reflog,
> doesn't it?  Is that a sensible assumption?

Oy. I forgot that local clock skew can cause this sort of behavior. :(

> Or does "the last log record" in the above mean "the log record with
> largest timestamp?  ... ah, not that is still not sufficient.  You'd
> still need to assume that timestamps on entries in one log block must
> be all older than the ones on entries in later log blocks.  Hmph...

Correct; I was assuming the times would be in order.

> Also it is not clear to me how reflogs for two refs would be
> intermixed in the log blocks, and what log keys for the entries must
> be recorded in the log index, to facilitate efficient lookup.  Is it
> assumed that a consecutive sequence of log blocks record reflogs for
> the same ref, before the sequence of log blocks switch to record
> reflogs for another ref, or something?

Multiple refs share the same log block.

>> A log index block must be written if 2 or more log blocks are written
>> to the file.  If present, the log index appears after the first log
>> block.  There is no padding used to align the log index to block
>> alignment.
>>
>> Log index format is identical to ref index, except the keys are 5
>> bytes longer to include `'\0'` and the 4-byte `reverse_int32(time)`.
>> Records use `block_offset` to refer to the start of a log block.
>
> I am assuming that we do not care about being able to quickly
> determine master@{24028}; I would imagine that it wouldn't be too
> hard to add an index to help such query, but I offhand would not
> know the details until I figure out how the format handles reflog
> entries for multiple refs first.

There is no assistance for master@{24028} quickly, its just a brute
force scan through 24,028 log records that pertain to master. Roughly
the same as the current reflog format.

Re: reftable [v2]: new ref storage format

[Ævar Arnfjörð Bjarmason]

On Tue, Jul 18 2017, Shawn Pearce jotted:

> On Mon, Jul 17, 2017 at 12:51 PM, Junio C Hamano <gitster@pobox.com> wrote:
>> Shawn Pearce <spearce@spearce.org> writes:
>>> where `time_sec` is the update time in seconds since the epoch.  The
>>> `reverse_int32` function inverses the value so lexographical ordering
>>> the network byte order time sorts more recent records first:
>>>
>>>     reverse_int(int32 t) {
>>>       return 0xffffffff - t;
>>>     }
>>
>> Is 2038 an issue, or by that time we'd all be retired together with
>> this file format and it won't be our problem?
>
> Based on discussion with Michael Haggerty, this is now an 8 byte field
> storing microseconds since the epoch. We should be good through year
> 9999.

I think this should be s/microseconds/nanoseconds/, not because there's
some great need to get better resolution than nanoseconds, but because:

 a) We already have WIP code (bp/fsmonitor) that's storing 64 bit
    nanoseconds since the epoch, albeit for the index, not for refs.

 b) There are several filesystems that have nanosecond resolution now,
    and it's likely more will start using that.

Thus:

 x) If you use such a filesystem you'll lose time resolution with this
    ref backend v.s. storing them on disk, which isn't itself a big
    deal, but more importantly you lose 1=1 time mapping as you
    transition and convert between the two.

 y) Our own code will need to juggle second resolution epochs
    (traditional FSs, any 32bit epoch format), microseconds (this
    proposal), and nanoseconds (new FSs, bp/fsmonitor) internally in
    various places.

    Let's not make this harder than it needs to be and just settle on
    two epoch resolution formats if we can help it, and so far it looks
    like we can.

The downside is that instead of lasting through the year 9999 the 64 bit
nanosecond resolution is only good up until the year 2554, which I think
is an acceptable trade-off given the above.

Re: reftable [v2]: new ref storage format

[Shawn Pearce]

My apologies for not responding to this piece of feedback earlier.

On Wed, Jul 19, 2017 at 7:02 AM, Ævar Arnfjörð Bjarmason
<avarab@gmail.com> wrote:
> On Tue, Jul 18 2017, Shawn Pearce jotted:
>> On Mon, Jul 17, 2017 at 12:51 PM, Junio C Hamano <gitster@pobox.com> wrote:
>>> Shawn Pearce <spearce@spearce.org> writes:
>>>> where `time_sec` is the update time in seconds since the epoch.  The
>>>> `reverse_int32` function inverses the value so lexographical ordering
>>>> the network byte order time sorts more recent records first:
>>>>
>>>>     reverse_int(int32 t) {
>>>>       return 0xffffffff - t;
>>>>     }
>>>
>>> Is 2038 an issue, or by that time we'd all be retired together with
>>> this file format and it won't be our problem?
>>
>> Based on discussion with Michael Haggerty, this is now an 8 byte field
>> storing microseconds since the epoch. We should be good through year
>> 9999.
>
> I think this should be s/microseconds/nanoseconds/, not because there's
> some great need to get better resolution than nanoseconds, but because:
>
>  a) We already have WIP code (bp/fsmonitor) that's storing 64 bit
>     nanoseconds since the epoch, albeit for the index, not for refs.
>
>  b) There are several filesystems that have nanosecond resolution now,
>     and it's likely more will start using that.

The time in a reflog and the time returned by lstat(2) to detect dirty
files in the working tree are unrelated. Of course we want the
dircache to be reflecting the highest precision available from lstat,
to reduce the number of files that must be content hashed for racily
clean detection. So if a filesystem is using nanoseconds, dircache
maybe should support it.

> Thus:
>
>  x) If you use such a filesystem you'll lose time resolution with this
>     ref backend v.s. storing them on disk, which isn't itself a big
>     deal, but more importantly you lose 1=1 time mapping as you
>     transition and convert between the two.

No, you won't. The reflog today ($GIT_DIR/logs) is storing second
precision in the log record. What precision the filesystem is using as
an mtime is irrelevant.

Further, microsecond is sufficient resolution for reflog data. From my
benchmarking just reading a reference from a very hot reftable costs
~20.2 usec. Any update of a reference requires a read-compare-modify
cycle, and so updates aren't going to be more frequent than 20 usec.

>  y) Our own code will need to juggle second resolution epochs
>     (traditional FSs, any 32bit epoch format), microseconds (this
>     proposal), and nanoseconds (new FSs, bp/fsmonitor) internally in
>     various places.

But these are also unrelated areas. IMHO, the nanosecond stuff should
be confined to the dircache management code and working tree
comparison code, and not be leaking out of there. Commit objects are
still recorded with second precision, and that isn't going to change.

Therefore I decided to stick with microseconds in the reftable v3
draft that I posted on July 22nd.

Re: reftable [v2]: new ref storage format

[Ævar Arnfjörð Bjarmason]

On Sun, Jul 23 2017, Shawn Pearce jotted:

> My apologies for not responding to this piece of feedback earlier.
>
> On Wed, Jul 19, 2017 at 7:02 AM, Ævar Arnfjörð Bjarmason
> <avarab@gmail.com> wrote:
>> On Tue, Jul 18 2017, Shawn Pearce jotted:
>>> On Mon, Jul 17, 2017 at 12:51 PM, Junio C Hamano <gitster@pobox.com> wrote:
>>>> Shawn Pearce <spearce@spearce.org> writes:
>>>>> where `time_sec` is the update time in seconds since the epoch.  The
>>>>> `reverse_int32` function inverses the value so lexographical ordering
>>>>> the network byte order time sorts more recent records first:
>>>>>
>>>>>     reverse_int(int32 t) {
>>>>>       return 0xffffffff - t;
>>>>>     }
>>>>
>>>> Is 2038 an issue, or by that time we'd all be retired together with
>>>> this file format and it won't be our problem?
>>>
>>> Based on discussion with Michael Haggerty, this is now an 8 byte field
>>> storing microseconds since the epoch. We should be good through year
>>> 9999.
>>
>> I think this should be s/microseconds/nanoseconds/, not because there's
>> some great need to get better resolution than nanoseconds, but because:
>>
>>  a) We already have WIP code (bp/fsmonitor) that's storing 64 bit
>>     nanoseconds since the epoch, albeit for the index, not for refs.
>>
>>  b) There are several filesystems that have nanosecond resolution now,
>>     and it's likely more will start using that.
>
> The time in a reflog and the time returned by lstat(2) to detect dirty
> files in the working tree are unrelated. Of course we want the
> dircache to be reflecting the highest precision available from lstat,
> to reduce the number of files that must be content hashed for racily
> clean detection. So if a filesystem is using nanoseconds, dircache
> maybe should support it.
>
>> Thus:
>>
>>  x) If you use such a filesystem you'll lose time resolution with this
>>     ref backend v.s. storing them on disk, which isn't itself a big
>>     deal, but more importantly you lose 1=1 time mapping as you
>>     transition and convert between the two.
>
> No, you won't. The reflog today ($GIT_DIR/logs) is storing second
> precision in the log record. What precision the filesystem is using as
> an mtime is irrelevant.

To this & the point above: Sorry about being unclear, I'm talking about
the mtime on the modified loose ref. This format proposes to replace
both loose & packed refs, does it not? The reflog time is not the only
place were we store the mtime of a ref. On my local ext4:

    $ tail -n 1 .git/logs/refs/heads/master
    <sha1> <sha1> Ævar Arnfjörð Bjarmason <avarab@gmail.com> 1500852355 +0200   commit: test
    $ perl -wE 'say ~~localtime shift' 1500852355
    Mon Jul 24 01:25:55 2017
    $ stat -c %y .git/logs/refs/heads/master
    2017-07-24 01:25:55.531379799 +0200

Of course you lose this information as soon as you "git pack-refs", but
it's there now & implicitly part of our current FS-backed on-disk
format.

So what I meant by "x" is that if to test this new reftable backend you
write a "git pack-reftable" you won't be able to 1=1 map it to the
mtimes you have on the fs showing when the ref was updated, but I see
now that you were perhaps never intending to use the more accurate FS
time at all for the loose refs, but just use the second resolution
reflog data.

> Further, microsecond is sufficient resolution for reflog data. From my
> benchmarking just reading a reference from a very hot reftable costs
> ~20.2 usec. Any update of a reference requires a read-compare-modify
> cycle, and so updates aren't going to be more frequent than 20 usec.

Right, I'm not arguing that it isn't sufficient, just that it's
introducing a needless variation by adding a third timestamp resolution
to git.

Even if it's not the same logical area in git (dir management v.s. ref
management) code to e.g. pretty format timestamps of sec/usec/nsec
resolution would tend to get shared, so we'd end up with 3 variants of
those instead of 2.

That's of course trivial, but so would be just deciding that ~500 years
of future proofing is good enough without any extra storage size for
those 64 bits and doing away with 1/3.

Just standardizing that makes more sense than picking the exact right
time resolution for every use case IMO. Otherwise we'll come up with
some other thingy in the future that just needs e.g. millisecond in its
format, and then end up with 4 variants....

I also see from "Update transactions" that unlike the current loose
backend the reftable backend wouldn't support multiple writers on
multiple machines (think NFS-mounted git master) updating unrelated
refs, which would break this usec assumption (but which holds due to the
locking involved in the new backend).

>>  y) Our own code will need to juggle second resolution epochs
>>     (traditional FSs, any 32bit epoch format), microseconds (this
>>     proposal), and nanoseconds (new FSs, bp/fsmonitor) internally in
>>     various places.
>
> But these are also unrelated areas. IMHO, the nanosecond stuff should
> be confined to the dircache management code and working tree
> comparison code, and not be leaking out of there. Commit objects are
> still recorded with second precision, and that isn't going to change.
>
> Therefore I decided to stick with microseconds in the reftable v3
> draft that I posted on July 22nd.