This PR replaces the existing Python-driven sync server with a new one in Rust.
The new server supports both collection and media syncing, and is compatible
with both the new protocol mentioned below, and older clients. A setting has
been added to the preferences screen to point Anki to a local server, and a
similar setting is likely to come to AnkiMobile soon.
Documentation is available here: <https://docs.ankiweb.net/sync-server.html>
In addition to the new server and refactoring, this PR also makes changes to the
sync protocol. The existing sync protocol places payloads and metadata inside a
multipart POST body, which causes a few headaches:
- Legacy clients build the request in a non-deterministic order, meaning the
entire request needs to be scanned to extract the metadata.
- Reqwest's multipart API directly writes the multipart body, without exposing
the resulting stream to us, making it harder to track the progress of the
transfer. We've been relying on a patched version of reqwest for timeouts,
which is a pain to keep up to date.
To address these issues, the metadata is now sent in a HTTP header, with the
data payload sent directly in the body. Instead of the slower gzip, we now
use zstd. The old timeout handling code has been replaced with a new implementation
that wraps the request and response body streams to track progress, allowing us
to drop the git dependencies for reqwest, hyper-timeout and tokio-io-timeout.
The main other change to the protocol is that one-way syncs no longer need to
downgrade the collection to schema 11 prior to sending.
* Relax chrono specification for AnkiDroid
https://github.com/ankidroid/Anki-Android-Backend/pull/251
* Add AnkiDroid service and AnkiDroid customizations
Most of the work here was done by David in the Backend repo; integrating
it into this repo for ease of future maintenance.
Based on 5d9f262f4c
with some tweaks:
- Protobuf imports have been fixed to match the recent refactor
- FatalError has been renamed to AnkidroidPanicError
- Tweaks to the desktop code to deal with the extra arg to open_collection,
and exclude AnkiDroid service methods from our Python code.
* Refactor AnkiDroid's DB code to avoid uses of unsafe
* Remove deprecated `and_hms()`
* Update chrono
* Update licenses and fix script
* Remove deprecated Date struct
* Remove chrono pin
* Skip format check on .vscode
Was failing for no reason.
* Replace deprecated chrono functions
* Add cargo-deny to update-licenses & pin versions (dae)
* Remove time 0.1 dependency (dae)
We don't need to wait for chrono 0.5; it was provided behind a legacy
feature flag.
(for upgrading users, please see the notes at the bottom)
Bazel brought a lot of nice things to the table, such as rebuilds based on
content changes instead of modification times, caching of build products,
detection of incorrect build rules via a sandbox, and so on. Rewriting the build
in Bazel was also an opportunity to improve on the Makefile-based build we had
prior, which was pretty poor: most dependencies were external or not pinned, and
the build graph was poorly defined and mostly serialized. It was not uncommon
for fresh checkouts to fail due to floating dependencies, or for things to break
when trying to switch to an older commit.
For day-to-day development, I think Bazel served us reasonably well - we could
generally switch between branches while being confident that builds would be
correct and reasonably fast, and not require full rebuilds (except on Windows,
where the lack of a sandbox and the TS rules would cause build breakages when TS
files were renamed/removed).
Bazel achieves that reliability by defining rules for each programming language
that define how source files should be turned into outputs. For the rules to
work with Bazel's sandboxing approach, they often have to reimplement or
partially bypass the standard tools that each programming language provides. The
Rust rules call Rust's compiler directly for example, instead of using Cargo,
and the Python rules extract each PyPi package into a separate folder that gets
added to sys.path.
These separate language rules allow proper declaration of inputs and outputs,
and offer some advantages such as caching of build products and fine-grained
dependency installation. But they also bring some downsides:
- The rules don't always support use-cases/platforms that the standard language
tools do, meaning they need to be patched to be used. I've had to contribute a
number of patches to the Rust, Python and JS rules to unblock various issues.
- The dependencies we use with each language sometimes make assumptions that do
not hold in Bazel, meaning they either need to be pinned or patched, or the
language rules need to be adjusted to accommodate them.
I was hopeful that after the initial setup work, things would be relatively
smooth-sailing. Unfortunately, that has not proved to be the case. Things
frequently broke when dependencies or the language rules were updated, and I
began to get frustrated at the amount of Anki development time I was instead
spending on build system upkeep. It's now about 2 years since switching to
Bazel, and I think it's time to cut losses, and switch to something else that's
a better fit.
The new build system is based on a small build tool called Ninja, and some
custom Rust code in build/. This means that to build Anki, Bazel is no longer
required, but Ninja and Rust need to be installed on your system. Python and
Node toolchains are automatically downloaded like in Bazel.
This new build system should result in faster builds in some cases:
- Because we're using cargo to build now, Rust builds are able to take advantage
of pipelining and incremental debug builds, which we didn't have with Bazel.
It's also easier to override the default linker on Linux/macOS, which can
further improve speeds.
- External Rust crates are now built with opt=1, which improves performance
of debug builds.
- Esbuild is now used to transpile TypeScript, instead of invoking the TypeScript
compiler. This results in faster builds, by deferring typechecking to test/check
time, and by allowing more work to happen in parallel.
As an example of the differences, when testing with the mold linker on Linux,
adding a new message to tags.proto (which triggers a recompile of the bulk of
the Rust and TypeScript code) results in a compile that goes from about 22s on
Bazel to about 7s in the new system. With the standard linker, it's about 9s.
Some other changes of note:
- Our Rust workspace now uses cargo-hakari to ensure all packages agree on
available features, preventing unnecessary rebuilds.
- pylib/anki is now a PEP420 implicit namespace, avoiding the need to merge
source files and generated files into a single folder for running. By telling
VSCode about the extra search path, code completion now works with generated
files without needing to symlink them into the source folder.
- qt/aqt can't use PEP420 as it's difficult to get rid of aqt/__init__.py.
Instead, the generated files are now placed in a separate _aqt package that's
added to the path.
- ts/lib is now exposed as @tslib, so the source code and generated code can be
provided under the same namespace without a merging step.
- MyPy and PyLint are now invoked once for the entire codebase.
- dprint will be used to format TypeScript/json files in the future instead of
the slower prettier (currently turned off to avoid causing conflicts). It can
automatically defer to prettier when formatting Svelte files.
- svelte-check is now used for typechecking our Svelte code, which revealed a
few typing issues that went undetected with the old system.
- The Jest unit tests now work on Windows as well.
If you're upgrading from Bazel, updated usage instructions are in docs/development.md and docs/build.md. A summary of the changes:
- please remove node_modules and .bazel
- install rustup (https://rustup.rs/)
- install rsync if not already installed (on windows, use pacman - see docs/windows.md)
- install Ninja (unzip from https://github.com/ninja-build/ninja/releases/tag/v1.11.1 and
place on your path, or from your distro/homebrew if it's 1.10+)
- update .vscode/settings.json from .vscode.dist
* Collection needs to be closed prior to backup even when not downgrading
* Backups -> BackupLimits
* Some improvements to backup_task
- backup_inner now returns the error instead of logging it, so that
the frontend can discover the issue when they await a backup (or create
another one)
- start_backup() was acquiring backup_task twice, and if another thread
started a backup between the two locks, the task could have been accidentally
overwritten without awaiting it
* Backups no longer require a collection close
- Instead of closing the collection, we ensure there is no active
transaction, and flush the WAL to disk. This means the undo history
is no longer lost on backup, which will be particularly useful if we
add a periodic backup in the future.
- Because a close is no longer required, backups are now achieved with
a separate command, instead of being included in CloseCollection().
- Full sync no longer requires an extra close+reopen step, and we now
wait for the backup to complete before proceeding.
- Create a backup before 'check db'
* Add File>Create Backup
https://forums.ankiweb.net/t/anki-mac-os-no-backup-on-sync/6157
* Defer checkpoint until we know we need it
When running periodic backups on a timer, we don't want to be fsync()ing
unnecessarily.
* Skip backup if modification time has not changed
We don't want the user leaving Anki open overnight, and coming back
to lots of identical backups.
* Periodic backups
Creates an automatic backup every 30 minutes if the collection has been
modified.
If there's a legacy checkpoint active, tries again 5 minutes later.
* Switch to a user-configurable backup duration
CreateBackup() now uses a simple force argument to determine whether
the user's limits should be respected or not, and only potentially
destructive ops (full download, check DB) override the user's configured
limit.
I considered having a separate limit for collection close and automatic
backups (eg keeping the previous 5 minute limit for collection close),
but that had two downsides:
- When the user closes their collection at the end of the day, they'd
get a recent backup. When they open the collection the next day, it
would get backed up again within 5 minutes, even though not much had
changed.
- Multiple limits are harder to communicate to users in the UI
Some remaining decisions I wasn't 100% sure about:
- If force is true but the collection has not been modified, the backup
will be skipped. If the user manually deleted their backups without
closing Anki, they wouldn't get a new one if the mtime hadn't changed.
- Force takes preference over the configured backup interval - should
we be ignored the user here, or take no backups at all?
Did a sneaky edit of the existing ftl string, as it hasn't been live
long.
* Move maybe_backup() into Collection
* Use a single method for manual and periodic backups
When manually creating a backup via the File menu, we no longer make
the user wait until the backup completes. As we continue waiting for
the backup in the background, if any errors occur, the user will get
notified about it fairly quickly.
* Show message to user if backup was skipped due to no changes
+ Don't incorrectly assert a backup will be created on force
* Add "automatic" to description
* Ensure we backup prior to importing colpkg if collection open
The backup doesn't happen when invoked from 'open backup' in the profile
screen, which matches Anki's previous behaviour. The user could
potentially clobber up to 30 minutes of their work if they exited to
the profile screen and restored a backup, but the alternative is we
create backups every time a backup is restored, which may happen a number
of times if the user is trying various ones. Or we could go back to a
separate throttle amount for this case, at the cost of more complexity.
* Remove the 0 special case on backup interval; minimum of 5 minutes
https://github.com/ankitects/anki/pull/1728#discussion_r830876833
* Write media files in chunks
* Test media file writing
* Add iter `ReadDirFiles`
* Remove ImportMediaError, fail fatally instead
Partially reverts commit e7cac9d1d6.
* Compare hashes of media files to be restored
* Improve `MediaCopier::copy()`
* Restore media files atomically with tempfile
* Make downgrade flag an enum
* Remove SchemaVersion::Latest in favour of Option
* Remove sha1 comparison again
* Remove unnecessary repr(u8) (dae)
* Avoid rebuilding regex in field search
* Special case search in all fields
* Don't repeat mid nodes in field search sql
Small speed gain for searches like `*:re:foo` and reduces the sql tree
depth if a lot of field names of the same notetype match.
* Add sql function to match fields with regex
* Optimise used field search algorithm
- Searching in all fields is a special case.
- Using native SQL comparison is preferred.
- For Regex, use newly added SQL function.
* Please clippy
* Avoid pyramid of doom
* nt_fields -> matched_fields
* Add tests for regex and all field searches
* minor tweaks for readability (dae)
This makes the review backlog case more expensive, since we end up
shuffling items outside the daily limit, but for the common case it's
about the same speed, and it means we don't need two separate sorting
steps. New cards remain handled the same way, since a backlog
is common there.
Also ensures that interday learning cards honor the deck sorting, and
that the non-default sort orders shuffle at the end.
So, this is fun. Apparently "DeckId" is considered preferable to the
"DeckID" were were using until now, and the latest clippy will start
warning about it. We could of course disable the warning, but probably
better to bite the bullet and switch to the naming that's generally
considered best.
Fixes the following issue:
- some code directly modifies the database, causing modified_in_python
to be set to true
- an undoable operation is run, which calls autosave() at the end
- autosave() notices there's an undoable operation, and commits immediately
- because modified_in_python was true, col.mtime was bumped in Python
- that invalidated the undo queue, preventing the operation from being
undone
This reverts commit 9008ba753e.
I fear this will be too disruptive - let's give AnkiDroid a bit more
time to catch up. Reverting this will mean new users are presented with
an upgrade notice on first startup, which looks a bit silly, but it's
probably the lesser of two evils.
- In corner cases, enabling the new timezone handling later can cause
reviews to shift forward or back a day, so it's best to have it on
by default.
- https://github.com/ankidroid/Anki-Android/issues/5805 has not landed
in a stable release yet, but will hopefully not be too far off by the
time 2.1.41 is released.
- Existing users will be unaffected, as the upgrade prompt in the previous
commit asks them if they use AnkiDroid.
- Users starting on AnkiDroid will be unaffected, as their collections
will still be on V1.
- The error message AnkiWeb gives when syncing an older AnkiDroid
with the new timezone enabled has been updated to direct users to the
preferences screen.
Came across a user with a corrupt index:
sqlite> pragma integrity_check;
integrity_check = wrong # of entries in index ix_revlog_cid
integrity_check = wrong # of entries in index ix_cards_sched
This is not picked up by a quick check, and a vacuum does not
fix it, but a reindex does.
AnkiWeb currently performs a full check when a file is uploaded
to it, so this was leading to "corrupt" to show up when syncing
the collection, with a local DB check not reporting/fixing the issue.
Closes#766
- changes the on-disk representation from % to a multiplier,
eg 250 -> 2.5, as this is consistent with the other options
- resets deck configs at or below 1.3 to 2.5
- for any cards that were using a reset deck config, reset their
current factor if it's at or below 2.0x. The cutoff is arbitrary,
and just intended to make sure we catch cards the user has rated
Easy on multiple times. The existing due dates are left alone.
- blue for normal sync, red for full sync required
- refactor status fetching code so we don't hold a collection lock
during the network request, which slows things down
- fix sync spinner restarting when returning to deck list
- notes with wrong field count are now recovered instead of
being deleted
- notes with missing note types are now recovered
- notes with missing cards are now recovered
- recover_missing_deck() still needs implementing
- checks required
