On Thu, Jun 7, 2018 at 7:03 AM Derrick Stolee <stolee@xxxxxxxxx> wrote: > > Signed-off-by: Derrick Stolee <dstolee@xxxxxxxxxxxxx> > --- > Documentation/technical/midx.txt | 109 +++++++++++++++++++++++++++++++ > 1 file changed, 109 insertions(+) > create mode 100644 Documentation/technical/midx.txt > > diff --git a/Documentation/technical/midx.txt b/Documentation/technical/midx.txt > new file mode 100644 > index 0000000000..789f410d71 > --- /dev/null > +++ b/Documentation/technical/midx.txt > @@ -0,0 +1,109 @@ > +Multi-Pack-Index (MIDX) Design Notes > +==================================== > + > +The Git object directory contains a 'pack' directory containing > +packfiles (with suffix ".pack") and pack-indexes (with suffix > +".idx"). The pack-indexes provide a way to lookup objects and > +navigate to their offset within the pack, but these must come > +in pairs with the packfiles. This pairing depends on the file > +names, as the pack-index differs only in suffix with its pack- > +file. While the pack-indexes provide fast lookup per packfile, > +this performance degrades as the number of packfiles increases, > +because abbreviations need to inspect every packfile and we are > +more likely to have a miss on our most-recently-used packfile. > +For some large repositories, repacking into a single packfile > +is not feasible due to storage space or excessive repack times. This leads to the question how MIDX will cope with large repos or a large number of packs. As it is just an index and not a pack itself, I guess it is smaller by some orders of magnitude, such that it is ok for now. > +The multi-pack-index (MIDX for short) stores a list of objects > +and their offsets into multiple packfiles. It contains: > + > +- A list of packfile names. > +- A sorted list of object IDs. > +- A list of metadata for the ith object ID including: > + - A value j referring to the jth packfile. > + - An offset within the jth packfile for the object. > +- If large offsets are required, we use another list of large > + offsets similar to version 2 pack-indexes. > + > +Thus, we can provide O(log N) lookup time for any number > +of packfiles. This sounds great for the lookup case! Though that is for the repo-read case. Let's read on how the dynamics of a repository are dealt with, e.g. integrating new packs into the MIDX, or how we deal with objects in multiple packs. > + > +Design Details > +-------------- > + > +- The MIDX is stored in a file named 'multi-pack-index' in the > + .git/objects/pack directory. This could be stored in the pack > + directory of an alternate. It refers only to packfiles in that > + same directory. So there is one and only one multi pack index? That makes the case of preparing the next MIDX that contains more pack references more interesting, as then we have to atomically update that file. > +- The core.midx config setting must be on to consume MIDX files. Looking through current config options, I would rename this to a more suggestive name. I searched for the core.idx counterpart that enables idx files -- it turns out that is named pack.indexVersion. So maybe pack.MultiIndex ? That could start out as a boolean as in this series and then evolve into a version number or such later. > +- The file format includes parameters for the object ID hash > + function, so a future change of hash algorithm does not require > + a change in format. > + > +- The MIDX keeps only one record per object ID. If an object appears > + in multiple packfiles, then the MIDX selects the copy in the most- > + recently modified packfile. Okay. That answers the question from above. Though this is just the tie breaking decision and not a hard limitation? (i.e. we could change this this later to that pack that has e.g. shortest delta chain for that object or such) > +- If there exist packfiles in the pack directory not registered in > + the MIDX, then those packfiles are loaded into the `packed_git` > + list and `packed_git_mru` cache. Not sure I understand the implications of this? Does that mean we first look at the multi index and if an object is not found, we'll search linearly through all packs that are not part of the MIDX? That would require the MIDX to be kepot up to date reasonably to be useful. > +- The pack-indexes (.idx files) remain in the pack directory so we > + can delete the MIDX file, set core.midx to false, or downgrade > + without any loss of information. In the future will it be possible to have no .idx files and just have the .midx? (I guess that depends on the strategy of how to integrate new packs into the MIDX?) > +- The MIDX file format uses a chunk-based approach (similar to the > + commit-graph file) that allows optional data to be added. ... or the index files v2 (or reftable files)? Sure, you are most familiar with commit-graph files, but others may find it easier to have some older file formats to relate to. > +Future Work > +----------- > + > +- Add a 'verify' subcommand to the 'git midx' builtin to verify the > + contents of the multi-pack-index file match the offsets listed in > + the corresponding pack-indexes. > + > +- The multi-pack-index allows many packfiles, especially in a context > + where repacking is expensive (such as a very large repo), or > + unexpected maintenance time is unacceptable (such as a high-demand > + build machine). Supposedly maintenance (git gc) can be run in the background without interfering with day-to-day life, how is the regeneration of commit graph or MIDX files impacting the work here? > However, the multi-pack-index needs to be rewritten > + in full every time. We can extend the format to be incremental, so > + writes are fast. By storing a small "tip" multi-pack-index that > + points to large "base" MIDX files, we can keep writes fast while > + still reducing the number of binary searches required for object > + lookups. So we can have multiple MIDX files? How would that work? Would there be a chunk that refers to other MIDX files? > +- The reachability bitmap is currently paired directly with a single > + packfile, using the pack-order as the object order to hopefully > + compress the bitmaps well using run-length encoding. This could be > + extended to pair a reachability bitmap with a multi-pack-index. If > + the multi-pack-index is extended to store a "stable object order" > + (a function Order(hash) = integer that is constant for a given hash, This stable object order doesn't fly well with integrating new packs? > + even as the multi-pack-index is updated) then a reachability bitmap > + could point to a multi-pack-index and be updated independently. > + > +- Packfiles can be marked as "special" using empty files that share > + the initial name but replace ".pack" with ".keep" or ".promisor". > + We can add an optional chunk of data to the multi-pack-index that > + records flags of information about the packfiles. This allows new > + states, such as 'repacked' or 'redeltified', that can help with > + pack maintenance in a multi-pack environment. It may also be > + helpful to organize packfiles by object type (commit, tree, blob, > + etc.) and use this metadata to help that maintenance. > + > +- The partial clone feature records special "promisor" packs that > + may point to objects that are not stored locally, but available > + on request to a server. The multi-pack-index does not currently > + track these promisor packs.
