syncthing/lib/db/db_test.go

702 lines
19 KiB
Go
Raw Permalink Normal View History

2014-11-16 20:13:20 +00:00
// Copyright (C) 2014 The Syncthing Authors.
//
2015-03-07 20:36:35 +00:00
// This Source Code Form is subject to the terms of the Mozilla Public
// License, v. 2.0. If a copy of the MPL was not distributed with this file,
// You can obtain one at https://mozilla.org/MPL/2.0/.
package db
import (
"bytes"
2020-07-11 07:22:15 +00:00
"context"
"fmt"
"testing"
"github.com/syncthing/syncthing/lib/db/backend"
"github.com/syncthing/syncthing/lib/events"
"github.com/syncthing/syncthing/lib/protocol"
)
func genBlocks(n int) []protocol.BlockInfo {
b := make([]protocol.BlockInfo, n)
for i := range b {
h := make([]byte, 32)
for j := range h {
h[j] = byte(i + j)
}
b[i].Size = i
b[i].Hash = h
}
return b
}
const myID = 1
var (
remoteDevice0, remoteDevice1 protocol.DeviceID
invalid = "invalid"
slashPrefixed = "/notgood"
haveUpdate0to3 map[protocol.DeviceID][]protocol.FileInfo
)
func init() {
remoteDevice0, _ = protocol.DeviceIDFromString("AIR6LPZ-7K4PTTV-UXQSMUU-CPQ5YWH-OEDFIIQ-JUG777G-2YQXXR5-YD6AWQR")
remoteDevice1, _ = protocol.DeviceIDFromString("I6KAH76-66SLLLB-5PFXSOA-UFJCDZC-YAOMLEK-CP2GB32-BV5RQST-3PSROAU")
haveUpdate0to3 = map[protocol.DeviceID][]protocol.FileInfo{
protocol.LocalDeviceID: {
protocol.FileInfo{Name: "a", Version: protocol.Vector{Counters: []protocol.Counter{{ID: myID, Value: 1000}}}, Blocks: genBlocks(1)},
protocol.FileInfo{Name: slashPrefixed, Version: protocol.Vector{Counters: []protocol.Counter{{ID: myID, Value: 1000}}}, Blocks: genBlocks(1)},
},
remoteDevice0: {
protocol.FileInfo{Name: "b", Version: protocol.Vector{Counters: []protocol.Counter{{ID: myID, Value: 1001}}}, Blocks: genBlocks(2)},
protocol.FileInfo{Name: "c", Version: protocol.Vector{Counters: []protocol.Counter{{ID: myID, Value: 1002}}}, Blocks: genBlocks(5), RawInvalid: true},
protocol.FileInfo{Name: "d", Version: protocol.Vector{Counters: []protocol.Counter{{ID: myID, Value: 1003}}}, Blocks: genBlocks(7)},
},
remoteDevice1: {
protocol.FileInfo{Name: "c", Version: protocol.Vector{Counters: []protocol.Counter{{ID: myID, Value: 1002}}}, Blocks: genBlocks(7)},
protocol.FileInfo{Name: "d", Version: protocol.Vector{Counters: []protocol.Counter{{ID: myID, Value: 1003}}}, Blocks: genBlocks(5), RawInvalid: true},
protocol.FileInfo{Name: invalid, Version: protocol.Vector{Counters: []protocol.Counter{{ID: myID, Value: 1004}}}, Blocks: genBlocks(5), RawInvalid: true},
},
}
}
// TestRepairSequence checks that a few hand-crafted messed-up sequence entries get fixed.
func TestRepairSequence(t *testing.T) {
db := newLowlevelMemory(t)
defer db.Close()
folderStr := "test"
folder := []byte(folderStr)
id := protocol.LocalDeviceID
short := protocol.LocalDeviceID.Short()
files := []protocol.FileInfo{
{Name: "fine", Blocks: genBlocks(1)},
{Name: "duplicate", Blocks: genBlocks(2)},
{Name: "missing", Blocks: genBlocks(3)},
{Name: "overwriting", Blocks: genBlocks(4)},
{Name: "inconsistent", Blocks: genBlocks(5)},
{Name: "inconsistentNotIndirected", Blocks: genBlocks(2)},
}
for i, f := range files {
files[i].Version = f.Version.Update(short)
}
trans, err := db.newReadWriteTransaction()
if err != nil {
t.Fatal(err)
}
defer trans.close()
addFile := func(f protocol.FileInfo, seq int64) {
dk, err := trans.keyer.GenerateDeviceFileKey(nil, folder, id[:], []byte(f.Name))
if err != nil {
t.Fatal(err)
}
if err := trans.putFile(dk, f); err != nil {
t.Fatal(err)
}
sk, err := trans.keyer.GenerateSequenceKey(nil, folder, seq)
if err != nil {
t.Fatal(err)
}
if err := trans.Put(sk, dk); err != nil {
t.Fatal(err)
}
}
// Plain normal entry
var seq int64 = 1
files[0].Sequence = 1
addFile(files[0], seq)
// Second entry once updated with original sequence still in place
f := files[1]
f.Sequence = int64(len(files) + 1)
addFile(f, f.Sequence)
// Original sequence entry
seq++
sk, err := trans.keyer.GenerateSequenceKey(nil, folder, seq)
if err != nil {
t.Fatal(err)
}
dk, err := trans.keyer.GenerateDeviceFileKey(nil, folder, id[:], []byte(f.Name))
if err != nil {
t.Fatal(err)
}
if err := trans.Put(sk, dk); err != nil {
t.Fatal(err)
}
// File later overwritten thus missing sequence entry
seq++
files[2].Sequence = seq
addFile(files[2], seq)
// File overwriting previous sequence entry (no seq bump)
seq++
files[3].Sequence = seq
addFile(files[3], seq)
// Inconistent files
seq++
files[4].Sequence = 101
addFile(files[4], seq)
seq++
files[5].Sequence = 102
addFile(files[5], seq)
// And a sequence entry pointing at nothing because why not
sk, err = trans.keyer.GenerateSequenceKey(nil, folder, 100001)
if err != nil {
t.Fatal(err)
}
dk, err = trans.keyer.GenerateDeviceFileKey(nil, folder, id[:], []byte("nonexisting"))
if err != nil {
t.Fatal(err)
}
if err := trans.Put(sk, dk); err != nil {
t.Fatal(err)
}
if err := trans.Commit(); err != nil {
t.Fatal(err)
}
// Loading the metadata for the first time means a "re"calculation happens,
// along which the sequences get repaired too.
db.gcMut.RLock()
_, err = db.loadMetadataTracker(folderStr)
db.gcMut.RUnlock()
if err != nil {
t.Fatal(err)
}
// Check the db
ro, err := db.newReadOnlyTransaction()
if err != nil {
t.Fatal(err)
}
defer ro.close()
it, err := ro.NewPrefixIterator([]byte{KeyTypeDevice})
if err != nil {
t.Fatal(err)
}
defer it.Release()
for it.Next() {
fi, err := ro.unmarshalTrunc(it.Value(), true)
if err != nil {
t.Fatal(err)
}
if sk, err = ro.keyer.GenerateSequenceKey(sk, folder, fi.SequenceNo()); err != nil {
t.Fatal(err)
}
dk, err := ro.Get(sk)
if backend.IsNotFound(err) {
t.Error("Missing sequence entry for", fi.FileName())
} else if err != nil {
t.Fatal(err)
}
if !bytes.Equal(it.Key(), dk) {
t.Errorf("Wrong key for %v, expected %s, got %s", f.FileName(), it.Key(), dk)
}
}
if err := it.Error(); err != nil {
t.Fatal(err)
}
it.Release()
it, err = ro.NewPrefixIterator([]byte{KeyTypeSequence})
if err != nil {
t.Fatal(err)
}
defer it.Release()
for it.Next() {
refactor: use modern Protobuf encoder (#9817) At a high level, this is what I've done and why: - I'm moving the protobuf generation for the `protocol`, `discovery` and `db` packages to the modern alternatives, and using `buf` to generate because it's nice and simple. - After trying various approaches on how to integrate the new types with the existing code, I opted for splitting off our own data model types from the on-the-wire generated types. This means we can have a `FileInfo` type with nicer ergonomics and lots of methods, while the protobuf generated type stays clean and close to the wire protocol. It does mean copying between the two when required, which certainly adds a small amount of inefficiency. If we want to walk this back in the future and use the raw generated type throughout, that's possible, this however makes the refactor smaller (!) as it doesn't change everything about the type for everyone at the same time. - I have simply removed in cold blood a significant number of old database migrations. These depended on previous generations of generated messages of various kinds and were annoying to support in the new fashion. The oldest supported database version now is the one from Syncthing 1.9.0 from Sep 7, 2020. - I changed config structs to be regular manually defined structs. For the sake of discussion, some things I tried that turned out not to work... ### Embedding / wrapping Embedding the protobuf generated structs in our existing types as a data container and keeping our methods and stuff: ``` package protocol type FileInfo struct { *generated.FileInfo } ``` This generates a lot of problems because the internal shape of the generated struct is quite different (different names, different types, more pointers), because initializing it doesn't work like you'd expect (i.e., you end up with an embedded nil pointer and a panic), and because the types of child types don't get wrapped. That is, even if we also have a similar wrapper around a `Vector`, that's not the type you get when accessing `someFileInfo.Version`, you get the `*generated.Vector` that doesn't have methods, etc. ### Aliasing ``` package protocol type FileInfo = generated.FileInfo ``` Doesn't help because you can't attach methods to it, plus all the above. ### Generating the types into the target package like we do now and attaching methods This fails because of the different shape of the generated type (as in the embedding case above) plus the generated struct already has a bunch of methods that we can't necessarily override properly (like `String()` and a bunch of getters). ### Methods to functions I considered just moving all the methods we attach to functions in a specific package, so that for example ``` package protocol func (f FileInfo) Equal(other FileInfo) bool ``` would become ``` package fileinfos func Equal(a, b *generated.FileInfo) bool ``` and this would mostly work, but becomes quite verbose and cumbersome, and somewhat limits discoverability (you can't see what methods are available on the type in auto completions, etc). In the end I did this in some cases, like in the database layer where a lot of things like `func (fv *FileVersion) IsEmpty() bool` becomes `func fvIsEmpty(fv *generated.FileVersion)` because they were anyway just internal methods. Fixes #8247
2024-12-01 15:50:17 +00:00
fi, ok, err := ro.getFileTrunc(it.Value(), false)
if err != nil {
t.Fatal(err)
}
seq := ro.keyer.SequenceFromSequenceKey(it.Key())
if !ok {
t.Errorf("Sequence entry %v points at nothing", seq)
} else if fi.SequenceNo() != seq {
t.Errorf("Inconsistent sequence entry for %v: %v != %v", fi.FileName(), fi.SequenceNo(), seq)
}
if len(fi.Blocks) == 0 {
t.Error("Missing blocks in", fi.FileName())
}
}
if err := it.Error(); err != nil {
t.Fatal(err)
}
it.Release()
}
func TestDowngrade(t *testing.T) {
db := newLowlevelMemory(t)
defer db.Close()
// sets the min version etc
if err := UpdateSchema(db); err != nil {
t.Fatal(err)
}
// Bump the database version to something newer than we actually support
miscDB := NewMiscDataNamespace(db)
if err := miscDB.PutInt64("dbVersion", dbVersion+1); err != nil {
t.Fatal(err)
}
l.Infoln(dbVersion)
// Pretend we just opened the DB and attempt to update it again
err := UpdateSchema(db)
if err, ok := err.(*databaseDowngradeError); !ok {
t.Fatal("Expected error due to database downgrade, got", err)
} else if err.minSyncthingVersion != dbMinSyncthingVersion {
t.Fatalf("Error has %v as min Syncthing version, expected %v", err.minSyncthingVersion, dbMinSyncthingVersion)
}
}
func TestCheckGlobals(t *testing.T) {
db := newLowlevelMemory(t)
defer db.Close()
fs := newFileSet(t, "test", db)
// Add any file
name := "foo"
fs.Update(protocol.LocalDeviceID, []protocol.FileInfo{
{
Name: name,
Type: protocol.FileInfoTypeFile,
Version: protocol.Vector{Counters: []protocol.Counter{{ID: 1, Value: 1001}}},
},
})
// Remove just the file entry
if err := db.dropPrefix([]byte{KeyTypeDevice}); err != nil {
t.Fatal(err)
}
// Clean up global entry of the now missing file
if repaired, err := db.checkGlobals(fs.folder); err != nil {
t.Fatal(err)
} else if repaired != 1 {
t.Error("Expected 1 repaired global item, got", repaired)
}
// Check that the global entry is gone
gk, err := db.keyer.GenerateGlobalVersionKey(nil, []byte(fs.folder), []byte(name))
if err != nil {
t.Fatal(err)
}
_, err = db.Get(gk)
if !backend.IsNotFound(err) {
t.Error("Expected key missing error, got", err)
}
}
func TestDropDuplicates(t *testing.T) {
names := []string{
"foo",
"bar",
"dcxvoijnds",
"3d/dsfase/4/ss2",
}
tcs := []struct{ in, out []int }{
{[]int{0}, []int{0}},
{[]int{0, 1}, []int{0, 1}},
{[]int{0, 1, 0, 1}, []int{0, 1}},
{[]int{0, 1, 1, 1, 1}, []int{0, 1}},
{[]int{0, 0, 0, 1}, []int{0, 1}},
{[]int{0, 1, 2, 3}, []int{0, 1, 2, 3}},
{[]int{3, 2, 1, 0, 0, 1, 2, 3}, []int{0, 1, 2, 3}},
{[]int{0, 1, 1, 3, 0, 1, 0, 1, 2, 3}, []int{0, 1, 2, 3}},
}
for tci, tc := range tcs {
inp := make([]protocol.FileInfo, len(tc.in))
expSeq := make(map[string]int)
for i, j := range tc.in {
inp[i] = protocol.FileInfo{Name: names[j], Sequence: int64(i)}
expSeq[names[j]] = i
}
outp := normalizeFilenamesAndDropDuplicates(inp)
if len(outp) != len(tc.out) {
t.Errorf("tc %v: Expected %v entries, got %v", tci, len(tc.out), len(outp))
continue
}
for i, f := range outp {
if exp := names[tc.out[i]]; exp != f.Name {
t.Errorf("tc %v: Got file %v at pos %v, expected %v", tci, f.Name, i, exp)
}
if exp := int64(expSeq[outp[i].Name]); exp != f.Sequence {
t.Errorf("tc %v: Got sequence %v at pos %v, expected %v", tci, f.Sequence, i, exp)
}
}
}
}
2020-07-11 07:22:15 +00:00
func TestGCIndirect(t *testing.T) {
// Verify that the gcIndirect run actually removes block lists.
db := newLowlevelMemory(t)
2020-07-11 07:22:15 +00:00
defer db.Close()
meta := newMetadataTracker(db.keyer, events.NoopLogger)
2020-07-11 07:22:15 +00:00
// Add three files with different block lists
files := []protocol.FileInfo{
{Name: "a", Blocks: genBlocks(100)},
{Name: "b", Blocks: genBlocks(200)},
{Name: "c", Blocks: genBlocks(300)},
}
db.updateLocalFiles([]byte("folder"), files, meta)
// Run a GC pass
db.gcIndirect(context.Background())
// Verify that we have three different block lists
n, err := numBlockLists(db)
if err != nil {
t.Fatal(err)
}
if n != len(files) {
t.Fatal("expected each file to have a block list")
}
// Change the block lists for each file
for i := range files {
files[i].Version = files[i].Version.Update(42)
files[i].Blocks = genBlocks(len(files[i].Blocks) + 1)
}
db.updateLocalFiles([]byte("folder"), files, meta)
// Verify that we now have *six* different block lists
n, err = numBlockLists(db)
if err != nil {
t.Fatal(err)
}
if n != 2*len(files) {
t.Fatal("expected both old and new block lists to exist")
}
// Run a GC pass
db.gcIndirect(context.Background())
// Verify that we now have just the three we need, again
n, err = numBlockLists(db)
if err != nil {
t.Fatal(err)
}
if n != len(files) {
t.Fatal("expected GC to collect all but the needed ones")
}
// Double check the correctness by loading the block lists and comparing with what we stored
tr, err := db.newReadOnlyTransaction()
if err != nil {
t.Fatal()
}
defer tr.Release()
for _, f := range files {
fi, ok, err := tr.getFile([]byte("folder"), protocol.LocalDeviceID[:], []byte(f.Name))
if err != nil {
t.Fatal(err)
}
if !ok {
t.Fatal("mysteriously missing")
}
if len(fi.Blocks) != len(f.Blocks) {
t.Fatal("block list mismatch")
}
for i := range fi.Blocks {
if !bytes.Equal(fi.Blocks[i].Hash, f.Blocks[i].Hash) {
t.Fatal("hash mismatch")
}
}
}
}
func TestUpdateTo14(t *testing.T) {
db := newLowlevelMemory(t)
defer db.Close()
folderStr := "default"
folder := []byte(folderStr)
name := []byte("foo")
file := protocol.FileInfo{Name: string(name), Version: protocol.Vector{Counters: []protocol.Counter{{ID: myID, Value: 1000}}}, Blocks: genBlocks(blocksIndirectionCutoff - 1)}
file.BlocksHash = protocol.BlocksHash(file.Blocks)
fileWOBlocks := file
fileWOBlocks.Blocks = nil
meta, err := db.loadMetadataTracker(folderStr)
if err != nil {
t.Fatal(err)
}
// Initially add the correct file the usual way, all good here.
if err := db.updateLocalFiles(folder, []protocol.FileInfo{file}, meta); err != nil {
t.Fatal(err)
}
// Simulate the previous bug, where .putFile could write a file info without
// blocks, even though the file has them (and thus a non-nil BlocksHash).
trans, err := db.newReadWriteTransaction()
if err != nil {
t.Fatal(err)
}
defer trans.close()
key, err := db.keyer.GenerateDeviceFileKey(nil, folder, protocol.LocalDeviceID[:], name)
if err != nil {
t.Fatal(err)
}
refactor: use modern Protobuf encoder (#9817) At a high level, this is what I've done and why: - I'm moving the protobuf generation for the `protocol`, `discovery` and `db` packages to the modern alternatives, and using `buf` to generate because it's nice and simple. - After trying various approaches on how to integrate the new types with the existing code, I opted for splitting off our own data model types from the on-the-wire generated types. This means we can have a `FileInfo` type with nicer ergonomics and lots of methods, while the protobuf generated type stays clean and close to the wire protocol. It does mean copying between the two when required, which certainly adds a small amount of inefficiency. If we want to walk this back in the future and use the raw generated type throughout, that's possible, this however makes the refactor smaller (!) as it doesn't change everything about the type for everyone at the same time. - I have simply removed in cold blood a significant number of old database migrations. These depended on previous generations of generated messages of various kinds and were annoying to support in the new fashion. The oldest supported database version now is the one from Syncthing 1.9.0 from Sep 7, 2020. - I changed config structs to be regular manually defined structs. For the sake of discussion, some things I tried that turned out not to work... ### Embedding / wrapping Embedding the protobuf generated structs in our existing types as a data container and keeping our methods and stuff: ``` package protocol type FileInfo struct { *generated.FileInfo } ``` This generates a lot of problems because the internal shape of the generated struct is quite different (different names, different types, more pointers), because initializing it doesn't work like you'd expect (i.e., you end up with an embedded nil pointer and a panic), and because the types of child types don't get wrapped. That is, even if we also have a similar wrapper around a `Vector`, that's not the type you get when accessing `someFileInfo.Version`, you get the `*generated.Vector` that doesn't have methods, etc. ### Aliasing ``` package protocol type FileInfo = generated.FileInfo ``` Doesn't help because you can't attach methods to it, plus all the above. ### Generating the types into the target package like we do now and attaching methods This fails because of the different shape of the generated type (as in the embedding case above) plus the generated struct already has a bunch of methods that we can't necessarily override properly (like `String()` and a bunch of getters). ### Methods to functions I considered just moving all the methods we attach to functions in a specific package, so that for example ``` package protocol func (f FileInfo) Equal(other FileInfo) bool ``` would become ``` package fileinfos func Equal(a, b *generated.FileInfo) bool ``` and this would mostly work, but becomes quite verbose and cumbersome, and somewhat limits discoverability (you can't see what methods are available on the type in auto completions, etc). In the end I did this in some cases, like in the database layer where a lot of things like `func (fv *FileVersion) IsEmpty() bool` becomes `func fvIsEmpty(fv *generated.FileVersion)` because they were anyway just internal methods. Fixes #8247
2024-12-01 15:50:17 +00:00
fiBs := mustMarshal(fileWOBlocks.ToWire(true))
if err := trans.Put(key, fiBs); err != nil {
t.Fatal(err)
}
if err := trans.Commit(); err != nil {
t.Fatal(err)
}
trans.close()
// Run migration, pretending were still on schema 13.
if err := (&schemaUpdater{db}).updateSchemaTo14(13); err != nil {
t.Fatal(err)
}
// checks
ro, err := db.newReadOnlyTransaction()
if err != nil {
t.Fatal(err)
}
defer ro.close()
if f, ok, err := ro.getFileByKey(key); err != nil {
t.Fatal(err)
} else if !ok {
t.Error("file missing")
} else if !f.MustRescan() {
t.Error("file not marked as MustRescan")
}
if vl, err := ro.getGlobalVersions(nil, folder, name); err != nil {
t.Fatal(err)
refactor: use modern Protobuf encoder (#9817) At a high level, this is what I've done and why: - I'm moving the protobuf generation for the `protocol`, `discovery` and `db` packages to the modern alternatives, and using `buf` to generate because it's nice and simple. - After trying various approaches on how to integrate the new types with the existing code, I opted for splitting off our own data model types from the on-the-wire generated types. This means we can have a `FileInfo` type with nicer ergonomics and lots of methods, while the protobuf generated type stays clean and close to the wire protocol. It does mean copying between the two when required, which certainly adds a small amount of inefficiency. If we want to walk this back in the future and use the raw generated type throughout, that's possible, this however makes the refactor smaller (!) as it doesn't change everything about the type for everyone at the same time. - I have simply removed in cold blood a significant number of old database migrations. These depended on previous generations of generated messages of various kinds and were annoying to support in the new fashion. The oldest supported database version now is the one from Syncthing 1.9.0 from Sep 7, 2020. - I changed config structs to be regular manually defined structs. For the sake of discussion, some things I tried that turned out not to work... ### Embedding / wrapping Embedding the protobuf generated structs in our existing types as a data container and keeping our methods and stuff: ``` package protocol type FileInfo struct { *generated.FileInfo } ``` This generates a lot of problems because the internal shape of the generated struct is quite different (different names, different types, more pointers), because initializing it doesn't work like you'd expect (i.e., you end up with an embedded nil pointer and a panic), and because the types of child types don't get wrapped. That is, even if we also have a similar wrapper around a `Vector`, that's not the type you get when accessing `someFileInfo.Version`, you get the `*generated.Vector` that doesn't have methods, etc. ### Aliasing ``` package protocol type FileInfo = generated.FileInfo ``` Doesn't help because you can't attach methods to it, plus all the above. ### Generating the types into the target package like we do now and attaching methods This fails because of the different shape of the generated type (as in the embedding case above) plus the generated struct already has a bunch of methods that we can't necessarily override properly (like `String()` and a bunch of getters). ### Methods to functions I considered just moving all the methods we attach to functions in a specific package, so that for example ``` package protocol func (f FileInfo) Equal(other FileInfo) bool ``` would become ``` package fileinfos func Equal(a, b *generated.FileInfo) bool ``` and this would mostly work, but becomes quite verbose and cumbersome, and somewhat limits discoverability (you can't see what methods are available on the type in auto completions, etc). In the end I did this in some cases, like in the database layer where a lot of things like `func (fv *FileVersion) IsEmpty() bool` becomes `func fvIsEmpty(fv *generated.FileVersion)` because they were anyway just internal methods. Fixes #8247
2024-12-01 15:50:17 +00:00
} else if fv, ok := vlGetGlobal(vl); !ok {
t.Error("missing global")
refactor: use modern Protobuf encoder (#9817) At a high level, this is what I've done and why: - I'm moving the protobuf generation for the `protocol`, `discovery` and `db` packages to the modern alternatives, and using `buf` to generate because it's nice and simple. - After trying various approaches on how to integrate the new types with the existing code, I opted for splitting off our own data model types from the on-the-wire generated types. This means we can have a `FileInfo` type with nicer ergonomics and lots of methods, while the protobuf generated type stays clean and close to the wire protocol. It does mean copying between the two when required, which certainly adds a small amount of inefficiency. If we want to walk this back in the future and use the raw generated type throughout, that's possible, this however makes the refactor smaller (!) as it doesn't change everything about the type for everyone at the same time. - I have simply removed in cold blood a significant number of old database migrations. These depended on previous generations of generated messages of various kinds and were annoying to support in the new fashion. The oldest supported database version now is the one from Syncthing 1.9.0 from Sep 7, 2020. - I changed config structs to be regular manually defined structs. For the sake of discussion, some things I tried that turned out not to work... ### Embedding / wrapping Embedding the protobuf generated structs in our existing types as a data container and keeping our methods and stuff: ``` package protocol type FileInfo struct { *generated.FileInfo } ``` This generates a lot of problems because the internal shape of the generated struct is quite different (different names, different types, more pointers), because initializing it doesn't work like you'd expect (i.e., you end up with an embedded nil pointer and a panic), and because the types of child types don't get wrapped. That is, even if we also have a similar wrapper around a `Vector`, that's not the type you get when accessing `someFileInfo.Version`, you get the `*generated.Vector` that doesn't have methods, etc. ### Aliasing ``` package protocol type FileInfo = generated.FileInfo ``` Doesn't help because you can't attach methods to it, plus all the above. ### Generating the types into the target package like we do now and attaching methods This fails because of the different shape of the generated type (as in the embedding case above) plus the generated struct already has a bunch of methods that we can't necessarily override properly (like `String()` and a bunch of getters). ### Methods to functions I considered just moving all the methods we attach to functions in a specific package, so that for example ``` package protocol func (f FileInfo) Equal(other FileInfo) bool ``` would become ``` package fileinfos func Equal(a, b *generated.FileInfo) bool ``` and this would mostly work, but becomes quite verbose and cumbersome, and somewhat limits discoverability (you can't see what methods are available on the type in auto completions, etc). In the end I did this in some cases, like in the database layer where a lot of things like `func (fv *FileVersion) IsEmpty() bool` becomes `func fvIsEmpty(fv *generated.FileVersion)` because they were anyway just internal methods. Fixes #8247
2024-12-01 15:50:17 +00:00
} else if !fvIsInvalid(fv) {
t.Error("global not marked as invalid")
}
}
func TestFlushRecursion(t *testing.T) {
// Verify that a commit hook can write to the transaction without
// causing another flush and thus recursion.
db := newLowlevelMemory(t)
defer db.Close()
// A commit hook that writes a small piece of data to the transaction.
hookFired := 0
hook := func(tx backend.WriteTransaction) error {
err := tx.Put([]byte(fmt.Sprintf("hook-key-%d", hookFired)), []byte(fmt.Sprintf("hook-value-%d", hookFired)))
if err != nil {
t.Fatal(err)
}
hookFired++
return nil
}
// A transaction.
tx, err := db.NewWriteTransaction(hook)
if err != nil {
t.Fatal(err)
}
defer tx.Release()
// Write stuff until the transaction flushes, thus firing the hook.
i := 0
for hookFired == 0 {
err := tx.Put([]byte(fmt.Sprintf("key-%d", i)), []byte(fmt.Sprintf("value-%d", i)))
if err != nil {
t.Fatal(err)
}
i++
}
// The hook should have fired precisely once.
if hookFired != 1 {
t.Error("expect one hook fire, not", hookFired)
}
}
func TestCheckLocalNeed(t *testing.T) {
db := newLowlevelMemory(t)
defer db.Close()
folderStr := "test"
fs := newFileSet(t, folderStr, db)
// Add files such that we are in sync for a and b, and need c and d.
files := []protocol.FileInfo{
{Name: "a", Version: protocol.Vector{Counters: []protocol.Counter{{ID: myID, Value: 1}}}},
{Name: "b", Version: protocol.Vector{Counters: []protocol.Counter{{ID: myID, Value: 1}}}},
{Name: "c", Version: protocol.Vector{Counters: []protocol.Counter{{ID: myID, Value: 1}}}},
{Name: "d", Version: protocol.Vector{Counters: []protocol.Counter{{ID: myID, Value: 1}}}},
}
fs.Update(protocol.LocalDeviceID, files)
files[2].Version = files[2].Version.Update(remoteDevice0.Short())
files[3].Version = files[2].Version.Update(remoteDevice0.Short())
fs.Update(remoteDevice0, files)
checkNeed := func() {
snap := snapshot(t, fs)
defer snap.Release()
c := snap.NeedSize(protocol.LocalDeviceID)
if c.Files != 2 {
t.Errorf("Expected 2 needed files locally, got %v in meta", c.Files)
}
needed := make([]protocol.FileInfo, 0, 2)
refactor: use modern Protobuf encoder (#9817) At a high level, this is what I've done and why: - I'm moving the protobuf generation for the `protocol`, `discovery` and `db` packages to the modern alternatives, and using `buf` to generate because it's nice and simple. - After trying various approaches on how to integrate the new types with the existing code, I opted for splitting off our own data model types from the on-the-wire generated types. This means we can have a `FileInfo` type with nicer ergonomics and lots of methods, while the protobuf generated type stays clean and close to the wire protocol. It does mean copying between the two when required, which certainly adds a small amount of inefficiency. If we want to walk this back in the future and use the raw generated type throughout, that's possible, this however makes the refactor smaller (!) as it doesn't change everything about the type for everyone at the same time. - I have simply removed in cold blood a significant number of old database migrations. These depended on previous generations of generated messages of various kinds and were annoying to support in the new fashion. The oldest supported database version now is the one from Syncthing 1.9.0 from Sep 7, 2020. - I changed config structs to be regular manually defined structs. For the sake of discussion, some things I tried that turned out not to work... ### Embedding / wrapping Embedding the protobuf generated structs in our existing types as a data container and keeping our methods and stuff: ``` package protocol type FileInfo struct { *generated.FileInfo } ``` This generates a lot of problems because the internal shape of the generated struct is quite different (different names, different types, more pointers), because initializing it doesn't work like you'd expect (i.e., you end up with an embedded nil pointer and a panic), and because the types of child types don't get wrapped. That is, even if we also have a similar wrapper around a `Vector`, that's not the type you get when accessing `someFileInfo.Version`, you get the `*generated.Vector` that doesn't have methods, etc. ### Aliasing ``` package protocol type FileInfo = generated.FileInfo ``` Doesn't help because you can't attach methods to it, plus all the above. ### Generating the types into the target package like we do now and attaching methods This fails because of the different shape of the generated type (as in the embedding case above) plus the generated struct already has a bunch of methods that we can't necessarily override properly (like `String()` and a bunch of getters). ### Methods to functions I considered just moving all the methods we attach to functions in a specific package, so that for example ``` package protocol func (f FileInfo) Equal(other FileInfo) bool ``` would become ``` package fileinfos func Equal(a, b *generated.FileInfo) bool ``` and this would mostly work, but becomes quite verbose and cumbersome, and somewhat limits discoverability (you can't see what methods are available on the type in auto completions, etc). In the end I did this in some cases, like in the database layer where a lot of things like `func (fv *FileVersion) IsEmpty() bool` becomes `func fvIsEmpty(fv *generated.FileVersion)` because they were anyway just internal methods. Fixes #8247
2024-12-01 15:50:17 +00:00
snap.WithNeed(protocol.LocalDeviceID, func(fi protocol.FileInfo) bool {
needed = append(needed, fi)
return true
})
if l := len(needed); l != 2 {
t.Errorf("Expected 2 needed files locally, got %v in db", l)
} else if needed[0].Name != "c" || needed[1].Name != "d" {
t.Errorf("Expected files c and d to be needed, got %v and %v", needed[0].Name, needed[1].Name)
}
}
checkNeed()
trans, err := db.newReadWriteTransaction()
if err != nil {
t.Fatal(err)
}
defer trans.close()
// Add "b" to needed and remove "d"
folder := []byte(folderStr)
key, err := trans.keyer.GenerateNeedFileKey(nil, folder, []byte(files[1].Name))
if err != nil {
t.Fatal(err)
}
if err = trans.Put(key, nil); err != nil {
t.Fatal(err)
}
key, err = trans.keyer.GenerateNeedFileKey(nil, folder, []byte(files[3].Name))
if err != nil {
t.Fatal(err)
}
if err = trans.Delete(key); err != nil {
t.Fatal(err)
}
if err := trans.Commit(); err != nil {
t.Fatal(err)
}
if repaired, err := db.checkLocalNeed(folder); err != nil {
t.Fatal(err)
} else if repaired != 2 {
t.Error("Expected 2 repaired local need items, got", repaired)
}
checkNeed()
}
func TestDuplicateNeedCount(t *testing.T) {
db := newLowlevelMemory(t)
defer db.Close()
folder := "test"
fs := newFileSet(t, folder, db)
files := []protocol.FileInfo{{Name: "foo", Version: protocol.Vector{}.Update(myID), Sequence: 1}}
fs.Update(protocol.LocalDeviceID, files)
files[0].Version = files[0].Version.Update(remoteDevice0.Short())
fs.Update(remoteDevice0, files)
db.checkRepair()
fs = newFileSet(t, folder, db)
found := false
for _, c := range fs.meta.counts.Counts {
refactor: use modern Protobuf encoder (#9817) At a high level, this is what I've done and why: - I'm moving the protobuf generation for the `protocol`, `discovery` and `db` packages to the modern alternatives, and using `buf` to generate because it's nice and simple. - After trying various approaches on how to integrate the new types with the existing code, I opted for splitting off our own data model types from the on-the-wire generated types. This means we can have a `FileInfo` type with nicer ergonomics and lots of methods, while the protobuf generated type stays clean and close to the wire protocol. It does mean copying between the two when required, which certainly adds a small amount of inefficiency. If we want to walk this back in the future and use the raw generated type throughout, that's possible, this however makes the refactor smaller (!) as it doesn't change everything about the type for everyone at the same time. - I have simply removed in cold blood a significant number of old database migrations. These depended on previous generations of generated messages of various kinds and were annoying to support in the new fashion. The oldest supported database version now is the one from Syncthing 1.9.0 from Sep 7, 2020. - I changed config structs to be regular manually defined structs. For the sake of discussion, some things I tried that turned out not to work... ### Embedding / wrapping Embedding the protobuf generated structs in our existing types as a data container and keeping our methods and stuff: ``` package protocol type FileInfo struct { *generated.FileInfo } ``` This generates a lot of problems because the internal shape of the generated struct is quite different (different names, different types, more pointers), because initializing it doesn't work like you'd expect (i.e., you end up with an embedded nil pointer and a panic), and because the types of child types don't get wrapped. That is, even if we also have a similar wrapper around a `Vector`, that's not the type you get when accessing `someFileInfo.Version`, you get the `*generated.Vector` that doesn't have methods, etc. ### Aliasing ``` package protocol type FileInfo = generated.FileInfo ``` Doesn't help because you can't attach methods to it, plus all the above. ### Generating the types into the target package like we do now and attaching methods This fails because of the different shape of the generated type (as in the embedding case above) plus the generated struct already has a bunch of methods that we can't necessarily override properly (like `String()` and a bunch of getters). ### Methods to functions I considered just moving all the methods we attach to functions in a specific package, so that for example ``` package protocol func (f FileInfo) Equal(other FileInfo) bool ``` would become ``` package fileinfos func Equal(a, b *generated.FileInfo) bool ``` and this would mostly work, but becomes quite verbose and cumbersome, and somewhat limits discoverability (you can't see what methods are available on the type in auto completions, etc). In the end I did this in some cases, like in the database layer where a lot of things like `func (fv *FileVersion) IsEmpty() bool` becomes `func fvIsEmpty(fv *generated.FileVersion)` because they were anyway just internal methods. Fixes #8247
2024-12-01 15:50:17 +00:00
if protocol.LocalDeviceID == c.DeviceID && c.LocalFlags == needFlag {
if found {
t.Fatal("second need count for local device encountered")
}
found = true
}
}
if !found {
t.Fatal("no need count for local device encountered")
}
}
func TestNeedAfterDropGlobal(t *testing.T) {
db := newLowlevelMemory(t)
defer db.Close()
folder := "test"
fs := newFileSet(t, folder, db)
// Initial:
// Three devices and a file "test": local has Version 1, remoteDevice0
// Version 2 and remoteDevice2 doesn't have it.
// All of them have "bar", just so the db knows about remoteDevice2.
files := []protocol.FileInfo{
{Name: "foo", Version: protocol.Vector{}.Update(myID), Sequence: 1},
{Name: "bar", Version: protocol.Vector{}.Update(myID), Sequence: 2},
}
fs.Update(protocol.LocalDeviceID, files)
files[0].Version = files[0].Version.Update(myID)
fs.Update(remoteDevice0, files)
fs.Update(remoteDevice1, files[1:])
// remoteDevice1 needs one file: test
snap := snapshot(t, fs)
c := snap.NeedSize(remoteDevice1)
if c.Files != 1 {
t.Errorf("Expected 1 needed files initially, got %v", c.Files)
}
snap.Release()
// Drop remoteDevice0, i.e. remove all their files from db.
// That changes the global file, which is now what local has.
fs.Drop(remoteDevice0)
// remoteDevice1 still needs test.
snap = snapshot(t, fs)
c = snap.NeedSize(remoteDevice1)
if c.Files != 1 {
t.Errorf("Expected still 1 needed files, got %v", c.Files)
}
snap.Release()
}
2020-07-11 07:22:15 +00:00
func numBlockLists(db *Lowlevel) (int, error) {
it, err := db.Backend.NewPrefixIterator([]byte{KeyTypeBlockList})
if err != nil {
return 0, err
}
defer it.Release()
n := 0
for it.Next() {
n++
}
if err := it.Error(); err != nil {
return 0, err
}
return n, nil
}