This adds a thin type that holds the state associated with the
leveldb.DB, leaving the huge Instance type more or less stateless. Also
moves some keying stuff into the DB package so that other packages need
not know the keying specifics.
(This does not, yet, fix the cmd/stindex program, in order to keep the
diff size down. Hence the keying constants are still exported.)
The problem here is that we would update the sequence index before
updating the FileInfos, which would result in a high sequence number
pointing to a low-sequence FileInfo. The index sender would pick up the
high sequence number, send the old file, and think everything was good.
On the receiving side the old file is a no-op and ignored. The file
remains out of sync until another update for it happens.
This fixes that by correcting the order of operations in the database
update: first we remove old sequence index entries, then we update the
FileInfos (which now don't have anything pointing to them) and then we
add the sequence indexes (which the index sender can see).
The other option is to add "proper" transactions where required at the
database layer. I actually have a branch for that, but it's literally
thousands of lines of diff and I'm putting that off for another day as
this solves the problem...
Adds a receive only folder type that does not send changes, and where the user can optionally revert local changes. Also changes some of the icons to make the three folder types distinguishable.
We have the invalid bit to indicate that a file isn't good. That's enough for remote devices. For ourselves, it would be good to know sometimes why the file isn't good - because it's an unsupported type, because it matches an ignore pattern, or because we detected the data is bad and we need to rescan it.
Or, and this is the main future reason for the PR, because it's a change detected on a receive only device. We will want something like the invalid flag for those changes, but marking them as invalid today means the scanner will rehash them. Hence something more fine grained is required.
This introduces a LocalFlags fields to the FileInfo where we can stash things that we care about locally. For example,
FlagLocalUnsupported = 1 << 0 // The kind is unsupported, e.g. symlinks on Windows
FlagLocalIgnored = 1 << 1 // Matches local ignore patterns
FlagLocalMustRescan = 1 << 2 // Doesn't match content on disk, must be rechecked fully
The LocalFlags fields isn't sent over the wire; instead the Invalid attribute is calculated based on the flags at index sending time. It's on the FileInfo anyway because that's what we serialize to database etc.
The actual Invalid flag should after this just be considered when building the global state and figuring out availability for remote devices. It is not used for local file index entries.
To optimize WithNeed, which is called for the local device whenever an index
update is received. No tracking for remote devices to conserve db space, as
WithNeed is only queried for completion.
This adds a couple of utilities for transporting databases in JSON and a
test to load a database and verify a couple of invalid bits. The test
itself is quite pointless at the moment, but it lays the groundwork for
testing the migration of this data in the next step (after the invalid
bit should be changed to local flags for local files).
When that happens we need to have a database in the old format already
there in order to be able to test the migration.
Instead of walking and unmarshalling the entire db and sorting the resulting
file infos by sequence, add store device keys by sequence number in the
database. Thus only the required file infos need be unmarshalled and are already
sorted by index.
When dropping delta index IDs due to upgrade, only drop our local one.
Previously, when dropping all of them, we would trigger a full send in
both directions on first connect after upgrade. Then the other side
would upgrade, doing the same thing. Net effect is full index data gets
sent twice in both directions.
With this change we just drop our local ID, meaning we will send our
full index on first connect after upgrade. When the other side upgrades,
they will do the same. This is a bit less cruel.
Unignored files are marked as conflicting while scanning, which is then resolved
in the subsequent pull. Automatically reconciles needed items on send-only
folders, if they do not actually differ except for internal metadata.
* lib/db: Don't panic on negative counts (fixes#4659)
So, negative counts should never happen and hence the original idea to
panic. However, this sucks as the panic will happen in a folder runner,
be automatically swallowed by suture, and the runner gets restarted but
now we are in a bad state. (Related: #4758)
At the time of writing the global list is somewhat in flux (we've
changed how ignored files are handled, invalid bits, etc.) and I think
that can cause unusual conditions here. Hence just fixing up the numbers
instead until the next full recount.
When scanner.Walk detects a change, it now returns the new file info as well as the old file info. It also finds deleted and ignored files while scanning.
Also directory deletions are now always committed to db after their children to prevent temporary failure on remote due to non-empty directory.
This keeps the data we need about sequence numbers and object counts
persistently in the database. The sizeTracker is expanded into a
metadataTracker than handled multiple folders, and the Counts struct is
made protobuf serializable. It gains a Sequence field to assist in
tracking that as well, and a collection of Counts become a CountsSet
(for serialization purposes).
The initial database scan is also a consistency check of the global
entries. This shouldn't strictly be necessary. Nonetheless I added a
created timestamp to the metadata and set a variable to compare against
that. When the time since the metadata creation is old enough, we drop
the metadata and rebuild from scratch like we used to, while also
consistency checking.
A new environment variable STCHECKDBEVERY can override this interval,
and for example be set to zero to force the check immediately.
GitHub-Pull-Request: https://github.com/syncthing/syncthing/pull/4547
LGTM: imsodin
This removes a significant, complex chunk of database code. The
"replace" operation walked both the old and new in lockstep and made the
relevant changes to make the new situation correct. But since delta
indexes we pretty much never need this - we just used replace to drop
the existing data and start over.
This makes that explicit and removes the complexity.
(This is one of those things that would be annoying to make case
insensitive, while the actual "drop and then insert" that we do is
easier.)
This is fairly well unit tested...
The one change to the tests is to cover the fact that previously replace
with something identical didn't bump the sequence number, while
obviously removing everything and re-inserting does. This is not
behavior we depend on anywhere.
GitHub-Pull-Request: https://github.com/syncthing/syncthing/pull/4500
LGTM: imsodin, AudriusButkevicius
So, when first implementing the database layer I added panics on every
unexpected error condition mostly to be sure to flush out bugs and
inconsistencies. Then it became sort of standard, and we don't seem to
have many bugs here any more so the panics are usually caused by things
like checksum errors on read. But it's not an optimal user experience to
crash all the time.
Here I've weeded out most of the panics, while retaining a few "can't
happen" ones like errors on marshalling and write that we really can't
recover from.
For the rest, I'm mostly treating any read error as "entry didn't
exist". This should mean we'll rescan the file and correct the info (if
scanning) or treat it as a new file and do conflict handling (when
pulling). In some cases things like our global stats may be slightly
incorrect until a restart, if a database entry goes suddenly missing
during runtime.
All in all, I think this makes us a bit more robust and friendly without
introducing too many risks for the user. If the database is truly toast,
probably many other things on the system will be toast as well...
GitHub-Pull-Request: https://github.com/syncthing/syncthing/pull/4118
Harmonize how we use batches in the model, using ProtoSize() to judge
the actual weight of the entire batch instead of estimating. Use smaller
batches in the block map - I think we might have though that batch.Len()
in the leveldb was the batch size in bytes, but it's actually number of
operations.
GitHub-Pull-Request: https://github.com/syncthing/syncthing/pull/4114
One more step on the path of the great refactoring. Touches rwfolder a
little bit since it uses the Lstat from fs as well, but mostly this is
just on the scanner as rwfolder is scheduled for a later refactor.
There are a couple of usages of fs.DefaultFilesystem that will in the
end become a filesystem injected from the top, but that comes later.
GitHub-Pull-Request: https://github.com/syncthing/syncthing/pull/4070
LGTM: AudriusButkevicius, imsodin
After this change,
- Symlinks on Windows are always unsupported. Sorry.
- Symlinks are always enabled on other platforms. They are just a small
file like anything else. There is no need to special case them. If you
don't want to sync some symlinks, ignore them.
- The protocol doesn't differentiate between different "types" of
symlinks. If that distinction ever does become relevant the individual
devices can figure it out by looking at the destination when they
create the link.
It's backwards compatible in that all the old symlink types are still
understood to be symlinks, and the new SYMLINK type is equivalent to the
old SYMLINK_UNKNOWN which was always a valid way to do it.
GitHub-Pull-Request: https://github.com/syncthing/syncthing/pull/3962
LGTM: AudriusButkevicius
Also tweaks the proto definitions:
- [packed=false] on the block_indexes field to retain compat with
v0.14.16 and earlier.
- Uses the vendored protobuf package in include paths.
And, "build.go setup" will install the vendored protoc-gen-gogofast.
This should ensure that a proto rebuild isn't so dependent on whatever
version of the compiler and package the developer has installed...
GitHub-Pull-Request: https://github.com/syncthing/syncthing/pull/3864
We used to consider deleted files & directories 128 bytes large. After
the delta indexes change a bug slipped in where deleted files would be
weighted according to their old non-deleted size. Both ways are
incorrect (but the latest change made it worse), as if there are more
files deleted than remaining data in the repo the needSize can be
greater than the globalSize, resulting in a negative completion
percentage.
This change makes it so that deleted items are zero bytes large, which
makes more sense. Instead we expose the number of files that we need to
delete as a separate field in the Completion() result, and hack the
percentage down to 95% complete if it was 100% complete but we need to
delete files. This latter part is sort of ugly, but necessary to give
the user some sort of feedback.
GitHub-Pull-Request: https://github.com/syncthing/syncthing/pull/3556
Furthermore:
1. Cleans configs received, migrates them as we receive them.
2. Clears indexes of devices we no longer share the folder with
GitHub-Pull-Request: https://github.com/syncthing/syncthing/pull/3478
This adds a new nanoseconds field to the FileInfo, populates it during
scans and sets the non-truncated time in Chtimes calls.
The actual file modification time is defined as modified_s seconds +
modified_ns nanoseconds. It's expected that the modified_ns field is <=
1e9 (that is, all whole seconds should go in the modified_s field) but
not really enforced. Given that it's an int32 the timestamp can be
adjusted += ~2.9 seconds by the modified_ns field...
GitHub-Pull-Request: https://github.com/syncthing/syncthing/pull/3431
The previous commit loosened the locking around database updates.
Apparently that was not fine - what happens is that parallell updates
to the same file for different devices stomp on each others updates to
the global index, leaving it missing one of the two devices.
This changes the BEP protocol to use protocol buffer serialization
instead of XDR, and therefore also the database format. The local
discovery protocol is also updated to be protocol buffer format.
GitHub-Pull-Request: https://github.com/syncthing/syncthing/pull/3276
LGTM: AudriusButkevicius
This adds a metric for "committed items" to the database instance that I
use in the test code, and a couple of tests that ensure that scans that
don't change anything also don't commit anything.
There was a case in the scanner where we set the invalid bit on files
that are ignored, even though they were already ignored and had the
invalid bit set. I had assumed this would result in an extra database
commit, but it was in fact filtered out by the Set... Anyway, I think we
can save some work on not pushing that change to the Set at all.
GitHub-Pull-Request: https://github.com/syncthing/syncthing/pull/3298
When doing prefix scans in the database, "foo" should not be considered
a prefix of "foo2". Instead, it should match "foo" exactly and also
strings with the prefix "foo/". This is more restrictive than what the
standard leveldb prefix scan does so we add some code to enforce it.
Also exposes the initialScanCompleted on the rwfolder for testing, and
change it to be a channel (so we can wait for it from another
goroutine). Otherwise we can't be sure when the initial scan has
completed, and we need to wait for that or it might pick up changes
we're doing at an unexpected time.
GitHub-Pull-Request: https://github.com/syncthing/syncthing/pull/3067
This happens automatically in the background anyway, and it can take a
long time on low powered devices at an inconvenient time. We just want
to get up and running as quickly as possible.
GitHub-Pull-Request: https://github.com/syncthing/syncthing/pull/3000
We're going to need the db.Instance to keep some state, and for that to
work we need the same one passed around everywhere. Hence this moves the
leveldb-specific file opening stuff into the db package and exports the
dbInstance type.
This implements a new debug/trace infrastructure based on a slightly
hacked up logger. Instead of the traditional "if debug { ... }" I've
rewritten the logger to have no-op Debugln and Debugf, unless debugging
has been enabled for a given "facility". The "facility" is just a
string, typically a package name.
This will be slightly slower than before; but not that much as it's
mostly a function call that returns immediately. For the cases where it
matters (the Debugln takes a hex.Dump() of something for example, and
it's not in a very occasional "if err != nil" branch) there is an
l.ShouldDebug(facility) that is fast enough to be used like the old "if
debug".
The point of all this is that we can now toggle debugging for the
various packages on and off at runtime. There's a new method
/rest/system/debug that can be POSTed a set of facilities to enable and
disable debug for, or GET from to get a list of facilities with
descriptions and their current debug status.
Similarly a /rest/system/log?since=... can grab the latest log entries,
up to 250 of them (hardcoded constant in main.go) plus the initial few.
Not implemented in this commit (but planned) is a simple debug GUI
available on /debug that shows the current log in an easily pasteable
format and has checkboxes to enable the various debug facilities.
The debug instructions to a user then becomes "visit this URL, check
these boxes, reproduce your problem, copy and paste the log". The actual
log viewer on the hypothetical /debug URL can poll regularly for new log
entries and this bypass the 250 line limit.
The existing STTRACE=foo variable is still obeyed and just sets the
start state of the system.
