0
0
Fork 0
mirror of https://github.com/matrix-org/dendrite synced 2024-12-15 14:33:43 +01:00
dendrite/roomserver/state/state.go

1103 lines
41 KiB
Go
Raw Normal View History

// Copyright 2017 Vector Creations Ltd
// Copyright 2018 New Vector Ltd
// Copyright 2019-2020 The Matrix.org Foundation C.I.C.
//
// Licensed under the Apache License, Version 2.0 (the "License");
// you may not use this file except in compliance with the License.
// You may obtain a copy of the License at
//
// http://www.apache.org/licenses/LICENSE-2.0
//
// Unless required by applicable law or agreed to in writing, software
// distributed under the License is distributed on an "AS IS" BASIS,
// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
// See the License for the specific language governing permissions and
// limitations under the License.
package state
import (
"context"
"fmt"
"sort"
"time"
"github.com/matrix-org/dendrite/roomserver/storage"
"github.com/matrix-org/util"
"github.com/prometheus/client_golang/prometheus"
"github.com/matrix-org/dendrite/roomserver/types"
"github.com/matrix-org/gomatrixserverlib"
)
type StateResolution struct {
db storage.Database
Roomserver/federation input refactor (#2104) * Put federation client functions into their own file * Look for missing auth events in RS input * Remove retrieveMissingAuthEvents from federation API * Logging * Sorta transplanted the code over * Use event origin failing all else * Don't get stuck on mutexes: * Add verifier * Don't mark state events with zero snapshot NID as not existing * Check missing state if not an outlier before storing the event * Reject instead of soft-fail, don't copy roominfo so much * Use synchronous contexts, limit time to fetch missing events * Clean up some commented out bits * Simplify `/send` endpoint significantly * Submit async * Report errors on sending to RS input * Set max payload in NATS to 16MB * Tweak metrics * Add `workerForRoom` for tidiness * Try skipping unmarshalling errors for RespMissingEvents * Track missing prev events separately to avoid calculating state when not possible * Tweak logic around checking missing state * Care about state when checking missing prev events * Don't check missing state for create events * Try that again * Handle create events better * Send create room events as new * Use given event kind when sending auth/state events * Revert "Use given event kind when sending auth/state events" This reverts commit 089d64d271b5fca8c104e1554711187420dbebca. * Only search for missing prev events or state for new events * Tweaks * We only have missing prev if we don't supply state * Room version tweaks * Allow async inputs again * Apply backpressure to consumers/synchronous requests to hopefully stop things being overwhelmed * Set timeouts on roomserver input tasks (need to decide what timeout makes sense) * Use work queue policy, deliver all on restart * Reduce chance of duplicates being sent by NATS * Limit the number of servers we attempt to reduce backpressure * Some review comment fixes * Tidy up a couple things * Don't limit servers, randomise order using map * Some context refactoring * Update gmsl * Don't resend create events * Set stateIDs length correctly or else the roomserver thinks there are missing events when there aren't * Exclude our own servername * Try backing off servers * Make excluding self behaviour optional * Exclude self from g_m_e * Update sytest-whitelist * Update consumers for the roomserver output stream * Remember to send outliers for state returned from /gme * Make full HTTP tests less upsetti * Remove 'If a device list update goes missing, the server resyncs on the next one' from the sytest blacklist * Remove debugging test * Fix blacklist again, remove unnecessary duplicate context * Clearer contexts, don't use background in case there's something happening there * Don't queue up events more than once in memory * Correctly identify create events when checking for state * Fill in gaps again in /gme code * Remove `AuthEventIDs` from `InputRoomEvent` * Remove stray field Co-authored-by: Kegan Dougal <kegan@matrix.org>
2022-01-27 15:29:14 +01:00
roomInfo *types.RoomInfo
events map[types.EventNID]*gomatrixserverlib.Event
}
Roomserver/federation input refactor (#2104) * Put federation client functions into their own file * Look for missing auth events in RS input * Remove retrieveMissingAuthEvents from federation API * Logging * Sorta transplanted the code over * Use event origin failing all else * Don't get stuck on mutexes: * Add verifier * Don't mark state events with zero snapshot NID as not existing * Check missing state if not an outlier before storing the event * Reject instead of soft-fail, don't copy roominfo so much * Use synchronous contexts, limit time to fetch missing events * Clean up some commented out bits * Simplify `/send` endpoint significantly * Submit async * Report errors on sending to RS input * Set max payload in NATS to 16MB * Tweak metrics * Add `workerForRoom` for tidiness * Try skipping unmarshalling errors for RespMissingEvents * Track missing prev events separately to avoid calculating state when not possible * Tweak logic around checking missing state * Care about state when checking missing prev events * Don't check missing state for create events * Try that again * Handle create events better * Send create room events as new * Use given event kind when sending auth/state events * Revert "Use given event kind when sending auth/state events" This reverts commit 089d64d271b5fca8c104e1554711187420dbebca. * Only search for missing prev events or state for new events * Tweaks * We only have missing prev if we don't supply state * Room version tweaks * Allow async inputs again * Apply backpressure to consumers/synchronous requests to hopefully stop things being overwhelmed * Set timeouts on roomserver input tasks (need to decide what timeout makes sense) * Use work queue policy, deliver all on restart * Reduce chance of duplicates being sent by NATS * Limit the number of servers we attempt to reduce backpressure * Some review comment fixes * Tidy up a couple things * Don't limit servers, randomise order using map * Some context refactoring * Update gmsl * Don't resend create events * Set stateIDs length correctly or else the roomserver thinks there are missing events when there aren't * Exclude our own servername * Try backing off servers * Make excluding self behaviour optional * Exclude self from g_m_e * Update sytest-whitelist * Update consumers for the roomserver output stream * Remember to send outliers for state returned from /gme * Make full HTTP tests less upsetti * Remove 'If a device list update goes missing, the server resyncs on the next one' from the sytest blacklist * Remove debugging test * Fix blacklist again, remove unnecessary duplicate context * Clearer contexts, don't use background in case there's something happening there * Don't queue up events more than once in memory * Correctly identify create events when checking for state * Fill in gaps again in /gme code * Remove `AuthEventIDs` from `InputRoomEvent` * Remove stray field Co-authored-by: Kegan Dougal <kegan@matrix.org>
2022-01-27 15:29:14 +01:00
func NewStateResolution(db storage.Database, roomInfo *types.RoomInfo) StateResolution {
return StateResolution{
db: db,
roomInfo: roomInfo,
events: make(map[types.EventNID]*gomatrixserverlib.Event),
}
}
// LoadStateAtSnapshot loads the full state of a room at a particular snapshot.
// This is typically the state before an event or the current state of a room.
// Returns a sorted list of state entries or an error if there was a problem talking to the database.
func (v *StateResolution) LoadStateAtSnapshot(
ctx context.Context, stateNID types.StateSnapshotNID,
) ([]types.StateEntry, error) {
stateBlockNIDLists, err := v.db.StateBlockNIDs(ctx, []types.StateSnapshotNID{stateNID})
if err != nil {
return nil, err
}
// We've asked for exactly one snapshot from the db so we should have exactly one entry in the result.
stateBlockNIDList := stateBlockNIDLists[0]
stateEntryLists, err := v.db.StateEntries(ctx, stateBlockNIDList.StateBlockNIDs)
if err != nil {
return nil, err
}
stateEntriesMap := stateEntryListMap(stateEntryLists)
// Combine all the state entries for this snapshot.
// The order of state block NIDs in the list tells us the order to combine them in.
var fullState []types.StateEntry
for _, stateBlockNID := range stateBlockNIDList.StateBlockNIDs {
entries, ok := stateEntriesMap.lookup(stateBlockNID)
if !ok {
// This should only get hit if the database is corrupt.
// It should be impossible for an event to reference a NID that doesn't exist
panic(fmt.Errorf("corrupt DB: Missing state block numeric ID %d", stateBlockNID))
}
fullState = append(fullState, entries...)
}
// Stable sort so that the most recent entry for each state key stays
// remains later in the list than the older entries for the same state key.
sort.Stable(stateEntryByStateKeySorter(fullState))
// Unique returns the last entry and hence the most recent entry for each state key.
fullState = fullState[:util.Unique(stateEntryByStateKeySorter(fullState))]
return fullState, nil
}
// LoadStateAtEvent loads the full state of a room before a particular event.
func (v *StateResolution) LoadStateAtEvent(
ctx context.Context, eventID string,
) ([]types.StateEntry, error) {
snapshotNID, err := v.db.SnapshotNIDFromEventID(ctx, eventID)
if err != nil {
return nil, fmt.Errorf("LoadStateAtEvent.SnapshotNIDFromEventID failed for event %s : %s", eventID, err)
}
if snapshotNID == 0 {
return nil, fmt.Errorf("LoadStateAtEvent.SnapshotNIDFromEventID(%s) returned 0 NID, was this event stored?", eventID)
}
stateEntries, err := v.LoadStateAtSnapshot(ctx, snapshotNID)
if err != nil {
return nil, err
}
return stateEntries, nil
}
// LoadCombinedStateAfterEvents loads a snapshot of the state after each of the events
// and combines those snapshots together into a single list. At this point it is
// possible to run into duplicate (type, state key) tuples.
func (v *StateResolution) LoadCombinedStateAfterEvents(
ctx context.Context, prevStates []types.StateAtEvent,
) ([]types.StateEntry, error) {
stateNIDs := make([]types.StateSnapshotNID, len(prevStates))
for i, state := range prevStates {
stateNIDs[i] = state.BeforeStateSnapshotNID
}
// Fetch the state snapshots for the state before the each prev event from the database.
// Deduplicate the IDs before passing them to the database.
// There could be duplicates because the events could be state events where
// the snapshot of the room state before them was the same.
stateBlockNIDLists, err := v.db.StateBlockNIDs(ctx, UniqueStateSnapshotNIDs(stateNIDs))
if err != nil {
return nil, fmt.Errorf("v.db.StateBlockNIDs: %w", err)
}
var stateBlockNIDs []types.StateBlockNID
for _, list := range stateBlockNIDLists {
stateBlockNIDs = append(stateBlockNIDs, list.StateBlockNIDs...)
}
// Fetch the state entries that will be combined to create the snapshots.
// Deduplicate the IDs before passing them to the database.
// There could be duplicates because a block of state entries could be reused by
// multiple snapshots.
stateEntryLists, err := v.db.StateEntries(ctx, uniqueStateBlockNIDs(stateBlockNIDs))
if err != nil {
return nil, fmt.Errorf("v.db.StateEntries: %w", err)
}
stateBlockNIDsMap := stateBlockNIDListMap(stateBlockNIDLists)
stateEntriesMap := stateEntryListMap(stateEntryLists)
// Combine the entries from all the snapshots of state after each prev event into a single list.
var combined []types.StateEntry
for _, prevState := range prevStates {
// Grab the list of state data NIDs for this snapshot.
stateBlockNIDs, ok := stateBlockNIDsMap.lookup(prevState.BeforeStateSnapshotNID)
if !ok {
// This should only get hit if the database is corrupt.
// It should be impossible for an event to reference a NID that doesn't exist
panic(fmt.Errorf("corrupt DB: Missing state snapshot numeric ID %d", prevState.BeforeStateSnapshotNID))
}
// Combine all the state entries for this snapshot.
// The order of state block NIDs in the list tells us the order to combine them in.
var fullState []types.StateEntry
for _, stateBlockNID := range stateBlockNIDs {
entries, ok := stateEntriesMap.lookup(stateBlockNID)
if !ok {
// This should only get hit if the database is corrupt.
// It should be impossible for an event to reference a NID that doesn't exist
panic(fmt.Errorf("corrupt DB: Missing state block numeric ID %d", stateBlockNID))
}
fullState = append(fullState, entries...)
}
if prevState.IsStateEvent() && !prevState.IsRejected {
// If the prev event was a state event then add an entry for the event itself
// so that we get the state after the event rather than the state before.
fullState = append(fullState, prevState.StateEntry)
}
// Stable sort so that the most recent entry for each state key stays
// remains later in the list than the older entries for the same state key.
sort.Stable(stateEntryByStateKeySorter(fullState))
// Unique returns the last entry and hence the most recent entry for each state key.
fullState = fullState[:util.Unique(stateEntryByStateKeySorter(fullState))]
// Add the full state for this StateSnapshotNID.
combined = append(combined, fullState...)
}
return combined, nil
}
// DifferenceBetweeenStateSnapshots works out which state entries have been added and removed between two snapshots.
func (v *StateResolution) DifferenceBetweeenStateSnapshots(
ctx context.Context, oldStateNID, newStateNID types.StateSnapshotNID,
) (removed, added []types.StateEntry, err error) {
if oldStateNID == newStateNID {
// If the snapshot NIDs are the same then nothing has changed
return nil, nil, nil
}
var oldEntries []types.StateEntry
var newEntries []types.StateEntry
if oldStateNID != 0 {
oldEntries, err = v.LoadStateAtSnapshot(ctx, oldStateNID)
if err != nil {
return nil, nil, err
}
}
if newStateNID != 0 {
newEntries, err = v.LoadStateAtSnapshot(ctx, newStateNID)
if err != nil {
return nil, nil, err
}
}
var oldI int
var newI int
for {
switch {
case oldI == len(oldEntries):
// We've reached the end of the old entries.
// The rest of the new list must have been newly added.
added = append(added, newEntries[newI:]...)
return
case newI == len(newEntries):
// We've reached the end of the new entries.
// The rest of the old list must be have been removed.
removed = append(removed, oldEntries[oldI:]...)
return
case oldEntries[oldI] == newEntries[newI]:
// The entry is in both lists so skip over it.
oldI++
newI++
case oldEntries[oldI].LessThan(newEntries[newI]):
// The lists are sorted so the old entry being less than the new entry means that it only appears in the old list.
removed = append(removed, oldEntries[oldI])
oldI++
default:
// Reaching the default case implies that the new entry is less than the old entry.
// Since the lists are sorted this means that it only appears in the new list.
added = append(added, newEntries[newI])
newI++
}
}
}
// LoadStateAtSnapshotForStringTuples loads the state for a list of event type and state key pairs at a snapshot.
// This is used when we only want to load a subset of the room state at a snapshot.
// If there is no entry for a given event type and state key pair then it will be discarded.
// This is typically the state before an event or the current state of a room.
// Returns a sorted list of state entries or an error if there was a problem talking to the database.
func (v *StateResolution) LoadStateAtSnapshotForStringTuples(
ctx context.Context,
stateNID types.StateSnapshotNID,
stateKeyTuples []gomatrixserverlib.StateKeyTuple,
) ([]types.StateEntry, error) {
numericTuples, err := v.stringTuplesToNumericTuples(ctx, stateKeyTuples)
if err != nil {
return nil, err
}
return v.loadStateAtSnapshotForNumericTuples(ctx, stateNID, numericTuples)
}
// stringTuplesToNumericTuples converts the string state key tuples into numeric IDs
// If there isn't a numeric ID for either the event type or the event state key then the tuple is discarded.
// Returns an error if there was a problem talking to the database.
func (v *StateResolution) stringTuplesToNumericTuples(
ctx context.Context,
stringTuples []gomatrixserverlib.StateKeyTuple,
) ([]types.StateKeyTuple, error) {
eventTypes := make([]string, len(stringTuples))
stateKeys := make([]string, len(stringTuples))
for i := range stringTuples {
eventTypes[i] = stringTuples[i].EventType
stateKeys[i] = stringTuples[i].StateKey
}
eventTypes = util.UniqueStrings(eventTypes)
eventTypeMap, err := v.db.EventTypeNIDs(ctx, eventTypes)
if err != nil {
return nil, err
}
stateKeys = util.UniqueStrings(stateKeys)
stateKeyMap, err := v.db.EventStateKeyNIDs(ctx, stateKeys)
if err != nil {
return nil, err
}
var result []types.StateKeyTuple
for _, stringTuple := range stringTuples {
var numericTuple types.StateKeyTuple
var ok1, ok2 bool
numericTuple.EventTypeNID, ok1 = eventTypeMap[stringTuple.EventType]
numericTuple.EventStateKeyNID, ok2 = stateKeyMap[stringTuple.StateKey]
// Discard the tuple if there wasn't a numeric ID for either the event type or the state key.
if ok1 && ok2 {
result = append(result, numericTuple)
}
}
return result, nil
}
// loadStateAtSnapshotForNumericTuples loads the state for a list of event type and state key pairs at a snapshot.
// This is used when we only want to load a subset of the room state at a snapshot.
// If there is no entry for a given event type and state key pair then it will be discarded.
// This is typically the state before an event or the current state of a room.
// Returns a sorted list of state entries or an error if there was a problem talking to the database.
func (v *StateResolution) loadStateAtSnapshotForNumericTuples(
ctx context.Context,
stateNID types.StateSnapshotNID,
stateKeyTuples []types.StateKeyTuple,
) ([]types.StateEntry, error) {
stateBlockNIDLists, err := v.db.StateBlockNIDs(ctx, []types.StateSnapshotNID{stateNID})
if err != nil {
return nil, err
}
// We've asked for exactly one snapshot from the db so we should have exactly one entry in the result.
stateBlockNIDList := stateBlockNIDLists[0]
stateEntryLists, err := v.db.StateEntriesForTuples(
ctx, stateBlockNIDList.StateBlockNIDs, stateKeyTuples,
)
if err != nil {
return nil, err
}
stateEntriesMap := stateEntryListMap(stateEntryLists)
// Combine all the state entries for this snapshot.
// The order of state block NIDs in the list tells us the order to combine them in.
var fullState []types.StateEntry
for _, stateBlockNID := range stateBlockNIDList.StateBlockNIDs {
entries, ok := stateEntriesMap.lookup(stateBlockNID)
if !ok {
// If the block is missing from the map it means that none of its entries matched a requested tuple.
// This can happen if the block doesn't contain an update for one of the requested tuples.
// If none of the requested tuples are in the block then it can be safely skipped.
continue
}
fullState = append(fullState, entries...)
}
// Stable sort so that the most recent entry for each state key stays
// remains later in the list than the older entries for the same state key.
sort.Stable(stateEntryByStateKeySorter(fullState))
// Unique returns the last entry and hence the most recent entry for each state key.
fullState = fullState[:util.Unique(stateEntryByStateKeySorter(fullState))]
return fullState, nil
}
// LoadStateAfterEventsForStringTuples loads the state for a list of event type
// and state key pairs after list of events.
// This is used when we only want to load a subset of the room state after a list of events.
// If there is no entry for a given event type and state key pair then it will be discarded.
// This is typically the state before an event.
// Returns a sorted list of state entries or an error if there was a problem talking to the database.
func (v *StateResolution) LoadStateAfterEventsForStringTuples(
ctx context.Context,
prevStates []types.StateAtEvent,
stateKeyTuples []gomatrixserverlib.StateKeyTuple,
) ([]types.StateEntry, error) {
numericTuples, err := v.stringTuplesToNumericTuples(ctx, stateKeyTuples)
if err != nil {
return nil, err
}
return v.loadStateAfterEventsForNumericTuples(ctx, prevStates, numericTuples)
}
func (v *StateResolution) loadStateAfterEventsForNumericTuples(
ctx context.Context,
prevStates []types.StateAtEvent,
stateKeyTuples []types.StateKeyTuple,
) ([]types.StateEntry, error) {
if len(prevStates) == 1 {
// Fast path for a single event.
prevState := prevStates[0]
result, err := v.loadStateAtSnapshotForNumericTuples(
ctx, prevState.BeforeStateSnapshotNID, stateKeyTuples,
)
if err != nil {
return nil, err
}
if prevState.IsStateEvent() {
// The result is current the state before the requested event.
// We want the state after the requested event.
// If the requested event was a state event then we need to
// update that key in the result.
// If the requested event wasn't a state event then the state after
// it is the same as the state before it.
set := false
for i := range result {
if result[i].StateKeyTuple == prevState.StateKeyTuple {
result[i] = prevState.StateEntry
set = true
}
}
if !set { // no previous state exists for this event: add new state
result = append(result, prevState.StateEntry)
}
}
return result, nil
}
// Slow path for more that one event.
// Load the entire state so that we can do conflict resolution if we need to.
// TODO: The are some optimistations we could do here:
// 1) We only need to do conflict resolution if there is a conflict in the
// requested tuples so we might try loading just those tuples and then
// checking for conflicts.
// 2) When there is a conflict we still only need to load the state
// needed to do conflict resolution which would save us having to load
// the full state.
// TODO: Add metrics for this as it could take a long time for big rooms
// with large conflicts.
fullState, _, _, err := v.calculateStateAfterManyEvents(ctx, v.roomInfo.RoomVersion, prevStates)
if err != nil {
return nil, err
}
// Sort the full state so we can use it as a map.
sort.Sort(stateEntrySorter(fullState))
// Filter the full state down to the required tuples.
var result []types.StateEntry
for _, tuple := range stateKeyTuples {
eventNID, ok := stateEntryMap(fullState).lookup(tuple)
if ok {
result = append(result, types.StateEntry{
StateKeyTuple: tuple,
EventNID: eventNID,
})
}
}
sort.Sort(stateEntrySorter(result))
return result, nil
}
var calculateStateDurations = prometheus.NewHistogramVec(
prometheus.HistogramOpts{
Namespace: "dendrite",
Subsystem: "roomserver",
Name: "calculate_state_duration_milliseconds",
Help: "How long it takes to calculate the state after a list of events",
Buckets: []float64{ // milliseconds
5, 10, 25, 50, 75, 100, 200, 300, 400, 500,
1000, 2000, 3000, 4000, 5000, 6000,
7000, 8000, 9000, 10000, 15000, 20000, 30000,
},
},
// Takes two labels:
// algorithm:
// The algorithm used to calculate the state or the step it failed on if it failed.
// Labels starting with "_" are used to indicate when the algorithm fails halfway.
// outcome:
// Whether the state was successfully calculated.
//
// The possible values for algorithm are:
// empty_state -> The list of events was empty so the state is empty.
// no_change -> The state hasn't changed.
// single_delta -> There was a single event added to the state in a way that can be encoded as a single delta
// full_state_no_conflicts -> We created a new copy of the full room state, but didn't enounter any conflicts
// while doing so.
// full_state_with_conflicts -> We created a new copy of the full room state and had to resolve conflicts to do so.
// _load_state_block_nids -> Failed loading the state block nids for a single previous state.
// _load_combined_state -> Failed to load the combined state.
// _resolve_conflicts -> Failed to resolve conflicts.
[]string{"algorithm", "outcome"},
)
var calculateStatePrevEventLength = prometheus.NewSummaryVec(
prometheus.SummaryOpts{
Namespace: "dendrite",
Subsystem: "roomserver",
Name: "calculate_state_prev_event_length",
Help: "The length of the list of events to calculate the state after",
},
[]string{"algorithm", "outcome"},
)
var calculateStateFullStateLength = prometheus.NewSummaryVec(
prometheus.SummaryOpts{
Namespace: "dendrite",
Subsystem: "roomserver",
Name: "calculate_state_full_state_length",
Help: "The length of the full room state.",
},
[]string{"algorithm", "outcome"},
)
var calculateStateConflictLength = prometheus.NewSummaryVec(
prometheus.SummaryOpts{
Namespace: "dendrite",
Subsystem: "roomserver",
Name: "calculate_state_conflict_state_length",
Help: "The length of the conflicted room state.",
},
[]string{"algorithm", "outcome"},
)
type calculateStateMetrics struct {
algorithm string
startTime time.Time
prevEventLength int
fullStateLength int
conflictLength int
}
func (c *calculateStateMetrics) stop(stateNID types.StateSnapshotNID, err error) (types.StateSnapshotNID, error) {
var outcome string
if err == nil {
outcome = "success"
} else {
outcome = "failure"
}
calculateStateDurations.WithLabelValues(c.algorithm, outcome).Observe(
float64(time.Since(c.startTime).Milliseconds()),
)
calculateStatePrevEventLength.WithLabelValues(c.algorithm, outcome).Observe(
float64(c.prevEventLength),
)
calculateStateFullStateLength.WithLabelValues(c.algorithm, outcome).Observe(
float64(c.fullStateLength),
)
calculateStateConflictLength.WithLabelValues(c.algorithm, outcome).Observe(
float64(c.conflictLength),
)
return stateNID, err
}
func init() {
prometheus.MustRegister(
calculateStateDurations, calculateStatePrevEventLength,
calculateStateFullStateLength, calculateStateConflictLength,
)
}
// CalculateAndStoreStateBeforeEvent calculates a snapshot of the state of a room before an event.
// Stores the snapshot of the state in the database.
// Returns a numeric ID for the snapshot of the state before the event.
func (v *StateResolution) CalculateAndStoreStateBeforeEvent(
ctx context.Context,
event *gomatrixserverlib.Event,
isRejected bool,
) (types.StateSnapshotNID, error) {
// Load the state at the prev events.
prevStates, err := v.db.StateAtEventIDs(ctx, event.PrevEventIDs())
if err != nil {
return 0, err
}
// The state before this event will be the state after the events that came before it.
return v.CalculateAndStoreStateAfterEvents(ctx, prevStates)
}
// CalculateAndStoreStateAfterEvents finds the room state after the given events.
// Stores the resulting state in the database and returns a numeric ID for that snapshot.
func (v *StateResolution) CalculateAndStoreStateAfterEvents(
ctx context.Context,
prevStates []types.StateAtEvent,
) (types.StateSnapshotNID, error) {
metrics := calculateStateMetrics{startTime: time.Now(), prevEventLength: len(prevStates)}
if len(prevStates) == 0 {
// 2) There weren't any prev_events for this event so the state is
// empty.
metrics.algorithm = "empty_state"
stateNID, err := v.db.AddState(ctx, v.roomInfo.RoomNID, nil, nil)
if err != nil {
err = fmt.Errorf("v.db.AddState: %w", err)
}
return metrics.stop(stateNID, err)
}
if len(prevStates) == 1 {
prevState := prevStates[0]
if prevState.EventStateKeyNID == 0 || prevState.IsRejected {
// 3) None of the previous events were state events and they all
// have the same state, so this event has exactly the same state
// as the previous events.
// This should be the internal case.
metrics.algorithm = "no_change"
return metrics.stop(prevState.BeforeStateSnapshotNID, nil)
}
// The previous event was a state event so we need to store a copy
// of the previous state updated with that event.
stateBlockNIDLists, err := v.db.StateBlockNIDs(
ctx, []types.StateSnapshotNID{prevState.BeforeStateSnapshotNID},
)
if err != nil {
metrics.algorithm = "_load_state_blocks"
return metrics.stop(0, fmt.Errorf("v.db.StateBlockNIDs: %w", err))
}
stateBlockNIDs := stateBlockNIDLists[0].StateBlockNIDs
if len(stateBlockNIDs) < maxStateBlockNIDs {
// 4) The number of state data blocks is small enough that we can just
// add the state event as a block of size one to the end of the blocks.
metrics.algorithm = "single_delta"
stateNID, err := v.db.AddState(
ctx, v.roomInfo.RoomNID, stateBlockNIDs, []types.StateEntry{prevState.StateEntry},
)
if err != nil {
err = fmt.Errorf("v.db.AddState: %w", err)
}
return metrics.stop(stateNID, err)
}
// If there are too many deltas then we need to calculate the full state
// So fall through to calculateAndStoreStateAfterManyEvents
}
stateNID, err := v.calculateAndStoreStateAfterManyEvents(ctx, v.roomInfo.RoomNID, prevStates, metrics)
if err != nil {
return 0, fmt.Errorf("v.calculateAndStoreStateAfterManyEvents: %w", err)
}
return stateNID, nil
}
// maxStateBlockNIDs is the maximum number of state data blocks to use to encode a snapshot of room state.
// Increasing this number means that we can encode more of the state changes as simple deltas which means that
// we need fewer entries in the state data table. However making this number bigger will increase the size of
// the rows in the state table itself and will require more index lookups when retrieving a snapshot.
// TODO: Tune this to get the right balance between size and lookup performance.
const maxStateBlockNIDs = 64
// calculateAndStoreStateAfterManyEvents finds the room state after the given events.
// This handles the slow path of calculateAndStoreStateAfterEvents for when there is more than one event.
// Stores the resulting state and returns a numeric ID for the snapshot.
func (v *StateResolution) calculateAndStoreStateAfterManyEvents(
ctx context.Context,
roomNID types.RoomNID,
prevStates []types.StateAtEvent,
metrics calculateStateMetrics,
) (types.StateSnapshotNID, error) {
state, algorithm, conflictLength, err :=
v.calculateStateAfterManyEvents(ctx, v.roomInfo.RoomVersion, prevStates)
metrics.algorithm = algorithm
if err != nil {
return metrics.stop(0, fmt.Errorf("v.calculateStateAfterManyEvents: %w", err))
}
// TODO: Check if we can encode the new state as a delta against the
// previous state.
metrics.conflictLength = conflictLength
metrics.fullStateLength = len(state)
return metrics.stop(v.db.AddState(ctx, roomNID, nil, state))
}
func (v *StateResolution) calculateStateAfterManyEvents(
ctx context.Context, roomVersion gomatrixserverlib.RoomVersion,
prevStates []types.StateAtEvent,
) (state []types.StateEntry, algorithm string, conflictLength int, err error) {
var combined []types.StateEntry
// Conflict resolution.
// First stage: load the state after each of the prev events.
combined, err = v.LoadCombinedStateAfterEvents(ctx, prevStates)
if err != nil {
err = fmt.Errorf("v.LoadCombinedStateAfterEvents: %w", err)
algorithm = "_load_combined_state"
return
}
// Collect all the entries with the same type and key together.
// We don't care about the order here because the conflict resolution
// algorithm doesn't depend on the order of the prev events.
// Remove duplicate entires.
combined = combined[:util.SortAndUnique(stateEntrySorter(combined))]
// Find the conflicts
conflicts := findDuplicateStateKeys(combined)
if len(conflicts) > 0 {
conflictLength = len(conflicts)
// 5) There are conflicting state events, for each conflict workout
// what the appropriate state event is.
// Work out which entries aren't conflicted.
var notConflicted []types.StateEntry
for _, entry := range combined {
if _, ok := stateEntryMap(conflicts).lookup(entry.StateKeyTuple); !ok {
notConflicted = append(notConflicted, entry)
}
}
var resolved []types.StateEntry
resolved, err = v.resolveConflicts(ctx, roomVersion, notConflicted, conflicts)
if err != nil {
err = fmt.Errorf("v.resolveConflits: %w", err)
algorithm = "_resolve_conflicts"
return
}
algorithm = "full_state_with_conflicts"
2020-03-30 10:51:45 +02:00
state = resolved[:util.SortAndUnique(stateEntrySorter(resolved))]
} else {
algorithm = "full_state_no_conflicts"
// 6) There weren't any conflicts
state = combined
}
return
}
func (v *StateResolution) resolveConflicts(
ctx context.Context, version gomatrixserverlib.RoomVersion,
notConflicted, conflicted []types.StateEntry,
) ([]types.StateEntry, error) {
stateResAlgo, err := version.StateResAlgorithm()
if err != nil {
return nil, err
}
switch stateResAlgo {
case gomatrixserverlib.StateResV1:
return v.resolveConflictsV1(ctx, notConflicted, conflicted)
case gomatrixserverlib.StateResV2:
return v.resolveConflictsV2(ctx, notConflicted, conflicted)
}
Federation for v3/v4 rooms (#954) * Update gomatrixserverlib * Default to room version 4 * Update gomatrixserverlib * Limit prev_events and auth_events * Fix auth_events, prev_events * Fix linter issues * Update gomatrixserverlib * Fix getState * Update sytest-whitelist * Squashed commit of the following: commit 067b87506357c996fd6ddb11271db9469ad4ce80 Author: Neil Alexander <neilalexander@users.noreply.github.com> Date: Fri Apr 3 14:29:06 2020 +0100 Invites v2 endpoint (#952) * Start converting v1 invite endpoint to v2 * Update gomatrixserverlib * Early federationsender code for sending invites * Sending invites sorta happens now * Populate invite request with stripped state * Remodel a bit, don't reflect received invites * Handle invite_room_state * Handle room versions a bit better * Update gomatrixserverlib * Tweak order in destinationQueue.next * Revert check in processMessage * Tweak federation sender destination queue code a bit * Add comments commit 955244c09298d0e6c870377dad3af2ffa1f5e578 Author: Ben B <benne@klimlive.de> Date: Fri Apr 3 12:40:50 2020 +0200 use custom http client instead of the http DefaultClient (#823) This commit replaces the default client from the http lib with a custom one. The previously used default client doesn't come with a timeout. This could cause unwanted locks. That solution chosen here creates a http client in the base component dendrite with a constant timeout of 30 seconds. If it should be necessary to overwrite this, we could include the timeout in the dendrite configuration. Here it would be a good idea to extend the type "Address" by a timeout and create an http client for each service. Closes #820 Signed-off-by: Benedikt Bongartz <benne@klimlive.de> Co-authored-by: Kegsay <kegan@matrix.org> * Update sytest-whitelist, sytest-blacklist * Update go.mod/go.sum * Add some error wrapping for debug * Add a NOTSPEC to common/events.go * Perform state resolution at send_join * Set default room version to v2 again * Tweak GetCapabilities * Add comments to ResolveConflictsAdhoc * Update sytest-blacklist * go mod tidy * Update sytest-whitelist, sytest-blacklist * Update versions * Updates from review comments * Update sytest-blacklist, sytest-whitelist * Check room versions compatible at make_join, add some comments, update gomatrixserverlib, other tweaks * Set default room version back to v2 * Update gomatrixserverlib, sytest-whitelist
2020-04-09 16:46:06 +02:00
return nil, fmt.Errorf("unsupported state resolution algorithm %v", stateResAlgo)
}
// resolveConflicts resolves a list of conflicted state entries. It takes two lists.
// The first is a list of all state entries that are not conflicted.
// The second is a list of all state entries that are conflicted
// A state entry is conflicted when there is more than one numeric event ID for the same state key tuple.
// Returns a list that combines the entries without conflicts with the result of state resolution for the entries with conflicts.
// The returned list is sorted by state key tuple.
// Returns an error if there was a problem talking to the database.
func (v *StateResolution) resolveConflictsV1(
ctx context.Context,
notConflicted, conflicted []types.StateEntry,
) ([]types.StateEntry, error) {
// Load the conflicted events
conflictedEvents, eventIDMap, err := v.loadStateEvents(ctx, conflicted)
if err != nil {
return nil, err
}
// Work out which auth events we need to load.
needed := gomatrixserverlib.StateNeededForAuth(conflictedEvents)
// Find the numeric IDs for the necessary state keys.
var neededStateKeys []string
neededStateKeys = append(neededStateKeys, needed.Member...)
neededStateKeys = append(neededStateKeys, needed.ThirdPartyInvite...)
stateKeyNIDMap, err := v.db.EventStateKeyNIDs(ctx, neededStateKeys)
if err != nil {
return nil, err
}
// Load the necessary auth events.
tuplesNeeded := v.stateKeyTuplesNeeded(stateKeyNIDMap, needed)
var authEntries []types.StateEntry
for _, tuple := range tuplesNeeded {
if eventNID, ok := stateEntryMap(notConflicted).lookup(tuple); ok {
authEntries = append(authEntries, types.StateEntry{
StateKeyTuple: tuple,
EventNID: eventNID,
})
}
}
authEvents, _, err := v.loadStateEvents(ctx, authEntries)
if err != nil {
return nil, err
}
// Resolve the conflicts.
resolvedEvents := gomatrixserverlib.ResolveStateConflicts(conflictedEvents, authEvents)
// Map from the full events back to numeric state entries.
for _, resolvedEvent := range resolvedEvents {
entry, ok := eventIDMap[resolvedEvent.EventID()]
if !ok {
panic(fmt.Errorf("missing state entry for event ID %q", resolvedEvent.EventID()))
}
notConflicted = append(notConflicted, entry)
}
// Sort the result so it can be searched.
sort.Sort(stateEntrySorter(notConflicted))
return notConflicted, nil
}
// resolveConflicts resolves a list of conflicted state entries. It takes two lists.
// The first is a list of all state entries that are not conflicted.
// The second is a list of all state entries that are conflicted
// A state entry is conflicted when there is more than one numeric event ID for the same state key tuple.
// Returns a list that combines the entries without conflicts with the result of state resolution for the entries with conflicts.
// The returned list is sorted by state key tuple.
// Returns an error if there was a problem talking to the database.
func (v *StateResolution) resolveConflictsV2(
ctx context.Context,
notConflicted, conflicted []types.StateEntry,
) ([]types.StateEntry, error) {
estimate := len(conflicted) + len(notConflicted)
eventIDMap := make(map[string]types.StateEntry, estimate)
// Load the conflicted events
conflictedEvents, conflictedEventMap, err := v.loadStateEvents(ctx, conflicted)
if err != nil {
return nil, err
}
for k, v := range conflictedEventMap {
eventIDMap[k] = v
}
// Load the non-conflicted events
nonConflictedEvents, nonConflictedEventMap, err := v.loadStateEvents(ctx, notConflicted)
if err != nil {
return nil, err
}
for k, v := range nonConflictedEventMap {
eventIDMap[k] = v
}
// For each conflicted event, we will add a new set of auth events. Auth
// events may be duplicated across these sets but that's OK.
authSets := make(map[string][]*gomatrixserverlib.Event, len(conflicted))
authEvents := make([]*gomatrixserverlib.Event, 0, estimate*3)
authDifference := make([]*gomatrixserverlib.Event, 0, estimate)
// For each conflicted event, let's try and get the needed auth events.
neededStateKeys := make([]string, 16)
authEntries := make([]types.StateEntry, 16)
for _, conflictedEvent := range conflictedEvents {
// Work out which auth events we need to load.
key := conflictedEvent.EventID()
needed := gomatrixserverlib.StateNeededForAuth([]*gomatrixserverlib.Event{conflictedEvent})
// Find the numeric IDs for the necessary state keys.
neededStateKeys = neededStateKeys[:0]
neededStateKeys = append(neededStateKeys, needed.Member...)
neededStateKeys = append(neededStateKeys, needed.ThirdPartyInvite...)
stateKeyNIDMap, err := v.db.EventStateKeyNIDs(ctx, neededStateKeys)
if err != nil {
return nil, err
}
// Load the necessary auth events.
tuplesNeeded := v.stateKeyTuplesNeeded(stateKeyNIDMap, needed)
authEntries = authEntries[:0]
for _, tuple := range tuplesNeeded {
if eventNID, ok := stateEntryMap(notConflicted).lookup(tuple); ok {
authEntries = append(authEntries, types.StateEntry{
StateKeyTuple: tuple,
EventNID: eventNID,
})
}
}
// Store the newly found auth events in the auth set for this event.
authSets[key], _, err = v.loadStateEvents(ctx, authEntries)
if err != nil {
return nil, err
}
authEvents = append(authEvents, authSets[key]...)
}
// This function helps us to work out whether an event exists in one of the
// auth sets.
isInAuthList := func(k string, event *gomatrixserverlib.Event) bool {
for _, e := range authSets[k] {
if e.EventID() == event.EventID() {
return true
}
}
return false
}
// This function works out if an event exists in all of the auth sets.
isInAllAuthLists := func(event *gomatrixserverlib.Event) bool {
found := true
for k := range authSets {
found = found && isInAuthList(k, event)
}
return found
}
// Look through all of the auth events that we've been given and work out if
// there are any events which don't appear in all of the auth sets. If they
// don't then we add them to the auth difference.
for _, event := range authEvents {
if !isInAllAuthLists(event) {
authDifference = append(authDifference, event)
}
}
// Resolve the conflicts.
resolvedEvents := gomatrixserverlib.ResolveStateConflictsV2(
conflictedEvents,
nonConflictedEvents,
authEvents,
authDifference,
)
// Map from the full events back to numeric state entries.
for _, resolvedEvent := range resolvedEvents {
entry, ok := eventIDMap[resolvedEvent.EventID()]
if !ok {
panic(fmt.Errorf("missing state entry for event ID %q", resolvedEvent.EventID()))
}
notConflicted = append(notConflicted, entry)
}
// Sort the result so it can be searched.
sort.Sort(stateEntrySorter(notConflicted))
return notConflicted, nil
}
// stateKeyTuplesNeeded works out which numeric state key tuples we need to authenticate some events.
func (v *StateResolution) stateKeyTuplesNeeded(stateKeyNIDMap map[string]types.EventStateKeyNID, stateNeeded gomatrixserverlib.StateNeeded) []types.StateKeyTuple {
var keyTuples []types.StateKeyTuple
if stateNeeded.Create {
keyTuples = append(keyTuples, types.StateKeyTuple{
EventTypeNID: types.MRoomCreateNID,
EventStateKeyNID: types.EmptyStateKeyNID,
})
}
if stateNeeded.PowerLevels {
keyTuples = append(keyTuples, types.StateKeyTuple{
EventTypeNID: types.MRoomPowerLevelsNID,
EventStateKeyNID: types.EmptyStateKeyNID,
})
}
if stateNeeded.JoinRules {
keyTuples = append(keyTuples, types.StateKeyTuple{
EventTypeNID: types.MRoomJoinRulesNID,
EventStateKeyNID: types.EmptyStateKeyNID,
})
}
for _, member := range stateNeeded.Member {
stateKeyNID, ok := stateKeyNIDMap[member]
if ok {
keyTuples = append(keyTuples, types.StateKeyTuple{
EventTypeNID: types.MRoomMemberNID,
EventStateKeyNID: stateKeyNID,
})
}
}
for _, token := range stateNeeded.ThirdPartyInvite {
stateKeyNID, ok := stateKeyNIDMap[token]
if ok {
keyTuples = append(keyTuples, types.StateKeyTuple{
EventTypeNID: types.MRoomThirdPartyInviteNID,
EventStateKeyNID: stateKeyNID,
})
}
}
return keyTuples
}
// loadStateEvents loads the matrix events for a list of state entries.
// Returns a list of state events in no particular order and a map from string event ID back to state entry.
// The map can be used to recover which numeric state entry a given event is for.
// Returns an error if there was a problem talking to the database.
func (v *StateResolution) loadStateEvents(
ctx context.Context, entries []types.StateEntry,
) ([]*gomatrixserverlib.Event, map[string]types.StateEntry, error) {
result := make([]*gomatrixserverlib.Event, 0, len(entries))
eventEntries := make([]types.StateEntry, 0, len(entries))
eventNIDs := make([]types.EventNID, 0, len(entries))
for _, entry := range entries {
if e, ok := v.events[entry.EventNID]; ok {
result = append(result, e)
} else {
eventEntries = append(eventEntries, entry)
eventNIDs = append(eventNIDs, entry.EventNID)
}
}
events, err := v.db.Events(ctx, eventNIDs)
if err != nil {
return nil, nil, err
}
eventIDMap := map[string]types.StateEntry{}
for _, entry := range eventEntries {
event, ok := eventMap(events).lookup(entry.EventNID)
if !ok {
panic(fmt.Errorf("corrupt DB: Missing event numeric ID %d", entry.EventNID))
}
result = append(result, event.Event)
eventIDMap[event.Event.EventID()] = entry
v.events[entry.EventNID] = event.Event
}
return result, eventIDMap, nil
}
// findDuplicateStateKeys finds the state entries where the state key tuple appears more than once in a sorted list.
// Returns a sorted list of those state entries.
func findDuplicateStateKeys(a []types.StateEntry) []types.StateEntry {
var result []types.StateEntry
// j is the starting index of a block of entries with the same state key tuple.
j := 0
for i := 1; i < len(a); i++ {
// Check if the state key tuple matches the start of the block
if a[j].StateKeyTuple != a[i].StateKeyTuple {
// If the state key tuple is different then we've reached the end of a block of duplicates.
// Check if the size of the block is bigger than one.
// If the size is one then there was only a single entry with that state key tuple so we don't add it to the result
if j+1 != i {
// Add the block to the result.
result = append(result, a[j:i]...)
}
// Start a new block for the next state key tuple.
j = i
}
}
// Check if the last block with the same state key tuple had more than one event in it.
if j+1 != len(a) {
result = append(result, a[j:]...)
}
return result
}
type stateEntrySorter []types.StateEntry
func (s stateEntrySorter) Len() int { return len(s) }
func (s stateEntrySorter) Less(i, j int) bool { return s[i].LessThan(s[j]) }
func (s stateEntrySorter) Swap(i, j int) { s[i], s[j] = s[j], s[i] }
type stateBlockNIDListMap []types.StateBlockNIDList
func (m stateBlockNIDListMap) lookup(stateNID types.StateSnapshotNID) (stateBlockNIDs []types.StateBlockNID, ok bool) {
list := []types.StateBlockNIDList(m)
i := sort.Search(len(list), func(i int) bool {
return list[i].StateSnapshotNID >= stateNID
})
if i < len(list) && list[i].StateSnapshotNID == stateNID {
ok = true
stateBlockNIDs = list[i].StateBlockNIDs
}
return
}
type stateEntryListMap []types.StateEntryList
func (m stateEntryListMap) lookup(stateBlockNID types.StateBlockNID) (stateEntries []types.StateEntry, ok bool) {
list := []types.StateEntryList(m)
i := sort.Search(len(list), func(i int) bool {
return list[i].StateBlockNID >= stateBlockNID
})
if i < len(list) && list[i].StateBlockNID == stateBlockNID {
ok = true
stateEntries = list[i].StateEntries
}
return
}
type stateEntryByStateKeySorter []types.StateEntry
func (s stateEntryByStateKeySorter) Len() int { return len(s) }
func (s stateEntryByStateKeySorter) Less(i, j int) bool {
return s[i].StateKeyTuple.LessThan(s[j].StateKeyTuple)
}
func (s stateEntryByStateKeySorter) Swap(i, j int) { s[i], s[j] = s[j], s[i] }
type stateNIDSorter []types.StateSnapshotNID
func (s stateNIDSorter) Len() int { return len(s) }
func (s stateNIDSorter) Less(i, j int) bool { return s[i] < s[j] }
func (s stateNIDSorter) Swap(i, j int) { s[i], s[j] = s[j], s[i] }
func UniqueStateSnapshotNIDs(nids []types.StateSnapshotNID) []types.StateSnapshotNID {
return nids[:util.SortAndUnique(stateNIDSorter(nids))]
}
type stateBlockNIDSorter []types.StateBlockNID
func (s stateBlockNIDSorter) Len() int { return len(s) }
func (s stateBlockNIDSorter) Less(i, j int) bool { return s[i] < s[j] }
func (s stateBlockNIDSorter) Swap(i, j int) { s[i], s[j] = s[j], s[i] }
func uniqueStateBlockNIDs(nids []types.StateBlockNID) []types.StateBlockNID {
return nids[:util.SortAndUnique(stateBlockNIDSorter(nids))]
}
// Map from event type, state key tuple to numeric event ID.
// Implemented using binary search on a sorted array.
type stateEntryMap []types.StateEntry
// lookup an entry in the event map.
func (m stateEntryMap) lookup(stateKey types.StateKeyTuple) (eventNID types.EventNID, ok bool) {
// Since the list is sorted we can implement this using binary search.
// This is faster than using a hash map.
// We don't have to worry about pathological cases because the keys are fixed
// size and are controlled by us.
list := []types.StateEntry(m)
i := sort.Search(len(list), func(i int) bool {
return !list[i].StateKeyTuple.LessThan(stateKey)
})
if i < len(list) && list[i].StateKeyTuple == stateKey {
ok = true
eventNID = list[i].EventNID
}
return
}
// Map from numeric event ID to event.
// Implemented using binary search on a sorted array.
type eventMap []types.Event
// lookup an entry in the event map.
func (m eventMap) lookup(eventNID types.EventNID) (event *types.Event, ok bool) {
// Since the list is sorted we can implement this using binary search.
// This is faster than using a hash map.
// We don't have to worry about pathological cases because the keys are fixed
// size are controlled by us.
list := []types.Event(m)
i := sort.Search(len(list), func(i int) bool {
return list[i].EventNID >= eventNID
})
if i < len(list) && list[i].EventNID == eventNID {
ok = true
event = &list[i]
}
return
}