| 1 |
// Copyright 2018 Google LLC. All Rights Reserved. |
| 2 |
// |
| 3 |
// Licensed under the Apache License, Version 2.0 (the "License"); |
| 4 |
// you may not use this file except in compliance with the License. |
| 5 |
// You may obtain a copy of the License at |
| 6 |
// |
| 7 |
// http://www.apache.org/licenses/LICENSE-2.0 |
| 8 |
// |
| 9 |
// Unless required by applicable law or agreed to in writing, software |
| 10 |
// distributed under the License is distributed on an "AS IS" BASIS, |
| 11 |
// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. |
| 12 |
// See the License for the specific language governing permissions and |
| 13 |
// limitations under the License. |
| 14 |
|
| 15 |
// Package core provides transport-agnostic implementation of Migrillian tool. |
| 16 |
package core |
| 17 |
|
| 18 |
import ( |
| 19 |
"context" |
| 20 |
"fmt" |
| 21 |
"math/rand" |
| 22 |
"strconv" |
| 23 |
"sync" |
| 24 |
"time" |
| 25 |
|
| 26 |
ct "github.com/google/certificate-transparency-go" |
| 27 |
"github.com/google/certificate-transparency-go/client" |
| 28 |
"github.com/google/certificate-transparency-go/scanner" |
| 29 |
"github.com/google/certificate-transparency-go/trillian/migrillian/configpb" |
| 30 |
"k8s.io/klog/v2" |
| 31 |
|
| 32 |
"github.com/google/trillian/monitoring" |
| 33 |
"github.com/google/trillian/util/clock" |
| 34 |
"github.com/google/trillian/util/election2" |
| 35 |
"github.com/transparency-dev/merkle/proof" |
| 36 |
"github.com/transparency-dev/merkle/rfc6962" |
| 37 |
) |
| 38 |
|
| 39 |
var ( |
| 40 |
metrics treeMetrics |
| 41 |
metricsOnce sync.Once |
| 42 |
) |
| 43 |
|
| 44 |
// treeMetrics holds metrics keyed by Tree ID. |
| 45 |
type treeMetrics struct { |
| 46 |
masterRuns monitoring.Counter |
| 47 |
masterCancels monitoring.Counter |
| 48 |
controllerStarts monitoring.Counter |
| 49 |
isMaster monitoring.Gauge |
| 50 |
entriesFetched monitoring.Counter |
| 51 |
entriesSeen monitoring.Counter |
| 52 |
entriesStored monitoring.Counter |
| 53 |
sthTimestamp monitoring.Gauge |
| 54 |
sthTreeSize monitoring.Gauge |
| 55 |
} |
| 56 |
|
| 57 |
// initMetrics creates metrics using the factory, if not yet created. |
| 58 |
func initMetrics(mf monitoring.MetricFactory) { |
| 59 |
const treeID = "tree_id" |
| 60 |
metricsOnce.Do(func() { |
| 61 |
metrics = treeMetrics{ |
| 62 |
masterRuns: mf.NewCounter("master_runs", "Number of mastership runs.", treeID), |
| 63 |
masterCancels: mf.NewCounter("master_cancels", "Number of unexpected mastership cancelations.", treeID), |
| 64 |
controllerStarts: mf.NewCounter("controller_starts", "Number of Controller (re-)starts.", treeID), |
| 65 |
isMaster: mf.NewGauge("is_master", "The instance is currently the master.", treeID), |
| 66 |
entriesFetched: mf.NewCounter("entries_fetched", "Entries fetched from the source log.", treeID), |
| 67 |
entriesSeen: mf.NewCounter("entries_seen", "Entries seen by the submitters.", treeID), |
| 68 |
entriesStored: mf.NewCounter("entries_stored", "Entries successfully submitted to Trillian.", treeID), |
| 69 |
sthTimestamp: mf.NewGauge("sth_timestamp", "Timestamp of the last seen STH.", treeID), |
| 70 |
sthTreeSize: mf.NewGauge("sth_tree_size", "Tree size of the last seen STH.", treeID), |
| 71 |
} |
| 72 |
}) |
| 73 |
} |
| 74 |
|
| 75 |
// Options holds configuration for a Controller. |
| 76 |
type Options struct { |
| 77 |
scanner.FetcherOptions |
| 78 |
Submitters int |
| 79 |
ChannelSize int |
| 80 |
NoConsistencyCheck bool |
| 81 |
StartDelay time.Duration |
| 82 |
StopAfter time.Duration |
| 83 |
} |
| 84 |
|
| 85 |
// OptionsFromConfig returns Options created from the passed in config. |
| 86 |
func OptionsFromConfig(cfg *configpb.MigrationConfig) Options { |
| 87 |
opts := Options{ |
| 88 |
FetcherOptions: scanner.FetcherOptions{ |
| 89 |
BatchSize: int(cfg.BatchSize), |
| 90 |
ParallelFetch: int(cfg.NumFetchers), |
| 91 |
StartIndex: cfg.StartIndex, |
| 92 |
EndIndex: cfg.EndIndex, |
| 93 |
Continuous: cfg.IsContinuous, |
| 94 |
}, |
| 95 |
Submitters: int(cfg.NumSubmitters), |
| 96 |
ChannelSize: int(cfg.ChannelSize), |
| 97 |
NoConsistencyCheck: cfg.NoConsistencyCheck, |
| 98 |
} |
| 99 |
if cfg.NumFetchers == 0 { |
| 100 |
opts.ParallelFetch = 1 |
| 101 |
} |
| 102 |
if cfg.NumSubmitters == 0 { |
| 103 |
opts.Submitters = 1 |
| 104 |
} |
| 105 |
return opts |
| 106 |
} |
| 107 |
|
| 108 |
// Controller coordinates migration from a CT log to a Trillian tree. |
| 109 |
type Controller struct { |
| 110 |
opts Options |
| 111 |
ctClient *client.LogClient |
| 112 |
plClient *PreorderedLogClient |
| 113 |
ef election2.Factory |
| 114 |
label string |
| 115 |
} |
| 116 |
|
| 117 |
// NewController creates a Controller configured by the passed in options, CT |
| 118 |
// and Trillian clients, and a master election factory. |
| 119 |
// |
| 120 |
// The passed in MetricFactory is used to create per-tree metrics, and it |
| 121 |
// should be the same for all instances. However, it is used only once. |
| 122 |
func NewController( |
| 123 |
opts Options, |
| 124 |
ctClient *client.LogClient, |
| 125 |
plClient *PreorderedLogClient, |
| 126 |
ef election2.Factory, |
| 127 |
mf monitoring.MetricFactory, |
| 128 |
) *Controller { |
| 129 |
initMetrics(mf) |
| 130 |
l := strconv.FormatInt(plClient.treeID, 10) |
| 131 |
return &Controller{opts: opts, ctClient: ctClient, plClient: plClient, ef: ef, label: l} |
| 132 |
} |
| 133 |
|
| 134 |
// RunWhenMasterWithRestarts calls RunWhenMaster, and, if the migration is |
| 135 |
// configured with continuous mode, restarts it whenever it returns. |
| 136 |
func (c *Controller) RunWhenMasterWithRestarts(ctx context.Context) { |
| 137 |
uri := c.ctClient.BaseURI() |
| 138 |
treeID := c.plClient.treeID |
| 139 |
for run := true; run; run = c.opts.Continuous && ctx.Err() == nil { |
| 140 |
klog.Infof("Starting migration Controller (%d<-%q)", treeID, uri) |
| 141 |
if err := c.RunWhenMaster(ctx); err != nil { |
| 142 |
klog.Errorf("Controller.RunWhenMaster(%d<-%q): %v", treeID, uri, err) |
| 143 |
continue |
| 144 |
} |
| 145 |
klog.Infof("Controller stopped (%d<-%q)", treeID, uri) |
| 146 |
} |
| 147 |
} |
| 148 |
|
| 149 |
// RunWhenMaster is a master-elected version of Run method. It executes Run |
| 150 |
// whenever this instance captures mastership of the tree ID. As soon as the |
| 151 |
// instance stops being the master, Run is canceled. The method returns if a |
| 152 |
// severe error occurs, the passed in context is canceled, or fetching is |
| 153 |
// completed (in non-Continuous mode). Releases mastership when terminates. |
| 154 |
func (c *Controller) RunWhenMaster(ctx context.Context) error { |
| 155 |
// Avoid thundering herd when starting multiple tasks on the same tree. |
| 156 |
if err := sleepRandom(ctx, 0, c.opts.StartDelay); err != nil { |
| 157 |
return err // The context has been canceled. |
| 158 |
} |
| 159 |
|
| 160 |
el, err := c.ef.NewElection(ctx, c.label) |
| 161 |
if err != nil { |
| 162 |
return err |
| 163 |
} |
| 164 |
metrics.isMaster.Set(0, c.label) |
| 165 |
defer func(ctx context.Context) { |
| 166 |
metrics.isMaster.Set(0, c.label) |
| 167 |
if err := el.Close(ctx); err != nil { |
| 168 |
klog.Warningf("%s: Election.Close(): %v", c.label, err) |
| 169 |
} |
| 170 |
}(ctx) |
| 171 |
|
| 172 |
for { |
| 173 |
if err := el.Await(ctx); err != nil { |
| 174 |
return err |
| 175 |
} |
| 176 |
metrics.isMaster.Set(1, c.label) |
| 177 |
|
| 178 |
mctx, err := el.WithMastership(ctx) |
| 179 |
if err != nil { |
| 180 |
return err |
| 181 |
} else if err := mctx.Err(); err != nil { |
| 182 |
return err |
| 183 |
} |
| 184 |
|
| 185 |
klog.Infof("%s: running as master", c.label) |
| 186 |
metrics.masterRuns.Inc(c.label) |
| 187 |
|
| 188 |
// Run while still master (or until an error). |
| 189 |
err = c.runWithRestarts(mctx) |
| 190 |
if ctx.Err() != nil { |
| 191 |
// We have been externally canceled, so return the current error (which |
| 192 |
// could be nil or a cancelation-related error). |
| 193 |
return err |
| 194 |
} else if mctx.Err() == nil { |
| 195 |
// We are still the master, so try to resign and emit the real error. |
| 196 |
if rerr := el.Resign(ctx); rerr != nil { |
| 197 |
klog.Errorf("%s: Election.Resign(): %v", c.label, rerr) |
| 198 |
} |
| 199 |
return err |
| 200 |
} |
| 201 |
|
| 202 |
// Otherwise the mastership has been canceled, retry. |
| 203 |
metrics.isMaster.Set(0, c.label) |
| 204 |
metrics.masterCancels.Inc(c.label) |
| 205 |
} |
| 206 |
} |
| 207 |
|
| 208 |
// runWithRestarts calls Run until it succeeds or the context is done, in |
| 209 |
// continuous mode. For non-continuous mode it is simply equivalent to Run. |
| 210 |
func (c *Controller) runWithRestarts(ctx context.Context) error { |
| 211 |
err := c.Run(ctx) |
| 212 |
if !c.opts.Continuous { |
| 213 |
return err |
| 214 |
} |
| 215 |
for err != nil && ctx.Err() == nil { |
| 216 |
klog.Errorf("%s: Controller.Run: %v", c.label, err) |
| 217 |
if slerr := sleepRandom(ctx, 0, c.opts.StartDelay); slerr == nil { |
| 218 |
err = c.Run(ctx) |
| 219 |
} |
| 220 |
} |
| 221 |
return ctx.Err() |
| 222 |
} |
| 223 |
|
| 224 |
// Run transfers CT log entries obtained via the CT log client to a Trillian |
| 225 |
// pre-ordered log via Trillian client. If Options.Continuous is true then the |
| 226 |
// migration process runs continuously trying to keep up with the target CT |
| 227 |
// log. Returns if an error occurs, the context is canceled, or all the entries |
| 228 |
// have been transferred (in non-Continuous mode). |
| 229 |
func (c *Controller) Run(ctx context.Context) error { |
| 230 |
metrics.controllerStarts.Inc(c.label) |
| 231 |
stopAfter := randDuration(c.opts.StopAfter, c.opts.StopAfter) |
| 232 |
start := time.Now() |
| 233 |
|
| 234 |
// Note: Non-continuous runs are not affected by StopAfter. |
| 235 |
pos, err := c.fetchTail(ctx, 0) |
| 236 |
if err != nil { |
| 237 |
return err |
| 238 |
} |
| 239 |
if !c.opts.Continuous { |
| 240 |
return nil |
| 241 |
} |
| 242 |
|
| 243 |
for stopAfter == 0 || time.Since(start) < stopAfter { |
| 244 |
// TODO(pavelkalinnikov): Integrate runWithRestarts here. |
| 245 |
next, err := c.fetchTail(ctx, pos) |
| 246 |
if err != nil { |
| 247 |
return err |
| 248 |
} |
| 249 |
if next == pos { |
| 250 |
// TODO(pavelkalinnikov): Pause with accordance to the rate of growth. |
| 251 |
// TODO(pavelkalinnikov): Make the duration configurable. |
| 252 |
if err := clock.SleepContext(ctx, 30*time.Second); err != nil { |
| 253 |
return err |
| 254 |
} |
| 255 |
} |
| 256 |
pos = next |
| 257 |
} |
| 258 |
|
| 259 |
return nil |
| 260 |
} |
| 261 |
|
| 262 |
// fetchTail transfers entries within the range specified in FetcherConfig, |
| 263 |
// with respect to the passed in minimal position to start from, and the |
| 264 |
// current tree size obtained from an STH. |
| 265 |
func (c *Controller) fetchTail(ctx context.Context, begin uint64) (uint64, error) { |
| 266 |
treeSize, rootHash, err := c.plClient.getRoot(ctx) |
| 267 |
if err != nil { |
| 268 |
return 0, err |
| 269 |
} |
| 270 |
|
| 271 |
fo := c.opts.FetcherOptions |
| 272 |
if fo.Continuous { // Ignore range parameters in continuous mode. |
| 273 |
fo.StartIndex, fo.EndIndex = int64(treeSize), 0 |
| 274 |
// Use non-continuous Fetcher, as we implement continuity in Controller. |
| 275 |
// TODO(pavelkalinnikov): Don't overload Fetcher's Continuous flag. |
| 276 |
fo.Continuous = false |
| 277 |
} else if fo.StartIndex < 0 { |
| 278 |
fo.StartIndex = int64(treeSize) |
| 279 |
} |
| 280 |
if int64(begin) > fo.StartIndex { |
| 281 |
fo.StartIndex = int64(begin) |
| 282 |
} |
| 283 |
klog.Infof("%s: fetching range [%d, %d)", c.label, fo.StartIndex, fo.EndIndex) |
| 284 |
|
| 285 |
fetcher := scanner.NewFetcher(c.ctClient, &fo) |
| 286 |
sth, err := fetcher.Prepare(ctx) |
| 287 |
if err != nil { |
| 288 |
return 0, err |
| 289 |
} |
| 290 |
metrics.sthTimestamp.Set(float64(sth.Timestamp), c.label) |
| 291 |
metrics.sthTreeSize.Set(float64(sth.TreeSize), c.label) |
| 292 |
if sth.TreeSize <= begin { |
| 293 |
return begin, nil |
| 294 |
} |
| 295 |
|
| 296 |
if err := c.verifyConsistency(ctx, treeSize, rootHash, sth); err != nil { |
| 297 |
return 0, err |
| 298 |
} |
| 299 |
|
| 300 |
var wg sync.WaitGroup |
| 301 |
batches := make(chan scanner.EntryBatch, c.opts.ChannelSize) |
| 302 |
cctx, cancel := context.WithCancel(ctx) |
| 303 |
defer cancel() |
| 304 |
|
| 305 |
for w, cnt := 0, c.opts.Submitters; w < cnt; w++ { |
| 306 |
wg.Add(1) |
| 307 |
go func() { |
| 308 |
defer wg.Done() |
| 309 |
if err := c.runSubmitter(cctx, batches); err != nil { |
| 310 |
klog.Errorf("%s: Stopping due to submitter error: %v", c.label, err) |
| 311 |
cancel() // Stop the other submitters and the Fetcher. |
| 312 |
} |
| 313 |
}() |
| 314 |
} |
| 315 |
|
| 316 |
handler := func(b scanner.EntryBatch) { |
| 317 |
metrics.entriesFetched.Add(float64(len(b.Entries)), c.label) |
| 318 |
select { |
| 319 |
case batches <- b: |
| 320 |
case <-cctx.Done(): // Avoid deadlock when shutting down. |
| 321 |
} |
| 322 |
} |
| 323 |
|
| 324 |
err = fetcher.Run(cctx, handler) |
| 325 |
close(batches) |
| 326 |
wg.Wait() |
| 327 |
if err != nil { |
| 328 |
return 0, err |
| 329 |
} |
| 330 |
// Run may have returned nil despite a cancel() call. |
| 331 |
if err := cctx.Err(); err != nil { |
| 332 |
return 0, fmt.Errorf("failed to fetch and submit the entire tail: %v", err) |
| 333 |
} |
| 334 |
return sth.TreeSize, nil |
| 335 |
} |
| 336 |
|
| 337 |
// verifyConsistency checks that the provided verified Trillian root is |
| 338 |
// consistent with the CT log's STH. |
| 339 |
func (c *Controller) verifyConsistency(ctx context.Context, treeSize uint64, rootHash []byte, sth *ct.SignedTreeHead) error { |
| 340 |
if treeSize == 0 { |
| 341 |
// Any head is consistent with empty root -- unnecessary to request empty proof. |
| 342 |
return nil |
| 343 |
} |
| 344 |
if c.opts.NoConsistencyCheck { |
| 345 |
klog.Warningf("%s: skipping consistency check", c.label) |
| 346 |
return nil |
| 347 |
} |
| 348 |
pf, err := c.ctClient.GetSTHConsistency(ctx, treeSize, sth.TreeSize) |
| 349 |
if err != nil { |
| 350 |
return err |
| 351 |
} |
| 352 |
return proof.VerifyConsistency(rfc6962.DefaultHasher, treeSize, sth.TreeSize, |
| 353 |
pf, rootHash, sth.SHA256RootHash[:]) |
| 354 |
} |
| 355 |
|
| 356 |
// runSubmitter obtains CT log entry batches from the controller's channel and |
| 357 |
// submits them through Trillian client. Returns when the channel is closed, or |
| 358 |
// the client returns a non-recoverable error (an example of a recoverable |
| 359 |
// error is when Trillian write quota is exceeded). |
| 360 |
func (c *Controller) runSubmitter(ctx context.Context, batches <-chan scanner.EntryBatch) error { |
| 361 |
for b := range batches { |
| 362 |
entries := float64(len(b.Entries)) |
| 363 |
metrics.entriesSeen.Add(entries, c.label) |
| 364 |
|
| 365 |
end := b.Start + int64(len(b.Entries)) |
| 366 |
if err := c.plClient.addSequencedLeaves(ctx, &b); err != nil { |
| 367 |
// addSequencedLeaves failed to submit entries despite retries. At this |
| 368 |
// point there is not much we can do. Seemingly the best strategy is to |
| 369 |
// shut down the Controller. |
| 370 |
return fmt.Errorf("failed to add batch [%d, %d): %v", b.Start, end, err) |
| 371 |
} |
| 372 |
klog.Infof("%s: added batch [%d, %d)", c.label, b.Start, end) |
| 373 |
metrics.entriesStored.Add(entries, c.label) |
| 374 |
} |
| 375 |
return nil |
| 376 |
} |
| 377 |
|
| 378 |
// sleepRandom sleeps for random duration in [base, base+spread). |
| 379 |
func sleepRandom(ctx context.Context, base, spread time.Duration) error { |
| 380 |
d := randDuration(base, spread) |
| 381 |
if d == 0 { |
| 382 |
return nil |
| 383 |
} |
| 384 |
return clock.SleepContext(ctx, d) |
| 385 |
} |
| 386 |
|
| 387 |
// randDuration returns a random duration in [base, base+spread). |
| 388 |
func randDuration(base, spread time.Duration) time.Duration { |
| 389 |
d := base |
| 390 |
if spread != 0 { |
| 391 |
d += time.Duration(rand.Int63n(int64(spread))) |
| 392 |
} |
| 393 |
return d |
| 394 |
} |