Files
releases/plugins/bundle/plugin.go
T
Stephan Renatus 516dd47dd1 runtime+storage: integrate disk storage
With this change, the disk backend (badger) becomes available for
use with the OPA runtime properly:

It can be configured using the `storage.disk` key in OPA's config
(see included documentation).

When enabled,
- any data or policies stored with OPA will persist over restarts
- per-query metrics related to disk usage are reported
- Prometheus metrics per storage operation are exported

The main intention behind this feature is to optimize memory usage:
OPA can now operate on more data than fits into the allotted memory
resources. It is NOT meant to be used as a primary source of truth:
there are no backup/restore or desaster recovery procedures -- you
MUST secure the means to restore the data stored with OPA's disk
storage by yourself.

See also #4014. Future improvements around bundle loading are
planned.

Some notes on details:

storage/disk: impose same locking regime used with inmem

With this setup, we'll ensure:

- there is only one open write txn at a time
- there are any number of open read txns at a time
- writes are blocked when reads are inflight
- during a commit (and triggers being run), no read txns can be created

This is to ensure the same atomic policy update semantics when using
'disk" as we have with "inmem". We're basically opting out of badger's
currency control and transactionality guarantees. This is because we
cannot piggy back on that to ensure the atomic update we want.

There might be other ways -- using subscribers, and blocking in some
other place -- but this one seems preferrable since it mirrors inmem.

Part of the problem is ErrTxnTooLarge, and committing and renewing
txns when it occurs: that, which is the prescribed solution to txns
growing too big, also means that reads can see half of the "logical"
transaction having been committed, while the rest is still getting
processed.

Another approach would have been using `WriteBatch`, but that won't
let us read from the batch, only apply Set and Delete operations.
We currently need to read (via an iterator) to figure out if we
need to delete keys to replace something in the store.  There is
no DropPrefix operation on the badger txn, or the WriteBatch API.

storage/disk: remove commit-and-renew-txn code for txn-too-big errors

This would break transactional guarantees we care about: while there
can be only one write transaction at a time, read transactions may
happen while a write txn is underway -- with this commit-and-reset
logic, those would read partial data.

Now, the error will be returned to the caller. The maximum txn size
depends on the size of memtables, and could be tweaked manually.
In general, the caller should try to push multiple smaller increments
of the data.

storage/disk: implement noop MakeDir

The MakeDir operation as implemented in the backend-agnostic storage
code has become an issue with the disk store: to write /foo/bar/baz,
we'd have to read /foo (among other subdirs), and that can be _much_
work for the disk backend. With inmem, it's cheap, so this wasn't
problematic before.

Some of the storage/disk/txn.go logic had to be adjusted to properly
do the MakeDir steps implicitly.

The index argument addition to patch() in storage/disk/txn.go was
necessary to keep the error messages conforming to the previous
code path: previously, conflicts (arrays indexed as objects) would
be surfaced in the MakeDir step, now it's entangled with the patch
calculation.

storage/disk: check ctx.Err() in List/Get operations

This won't abort reading a single key, but it will abort iterations.

storage/disk: support patterns in partitions

There is a potential clash here: "*", the path wildcard, is
a valid path section. However, it only affects the case when
a user would want to have a partition at

    /foo/*/bar

and would really mean "*", and not the wildcard.

Storing data at /foo/*/bar with a literal "*" won't be treated
differently than storing something at /fo/xyz/bar.

storage/disk: keep per-txn-type histograms of stats

This is done by reading off the metrics on commit, and shovelling
their numbers into the prometheus collector.

NOTE: if you were to share a metrics object among multiple transactions,
the results would be skewed, as it's not reset. However, our server
handlers don't do that.

storage/disk: opt out of badger's conflict detection

With only one write transaction in flight at any time, the situation
that badger guards against cannot happen:

A transaction has written to a key after the current, to-be-committed
transaction has last read that key from the store.

Since it can't happen, we can ignore the bookkeeping involved. This
improves the time it takes to overwrite existing keys.

Signed-off-by: Stephan Renatus <stephan.renatus@gmail.com>
2022-03-30 10:25:45 +02:00

760 lines
20 KiB
Go

// Copyright 2018 The OPA Authors. All rights reserved.
// Use of this source code is governed by an Apache2
// license that can be found in the LICENSE file.
// Package bundle implements bundle loading.
package bundle
import (
"context"
"errors"
"fmt"
"io"
"net/url"
"os"
"path/filepath"
"reflect"
"sync"
"time"
"github.com/open-policy-agent/opa/ast"
"github.com/open-policy-agent/opa/bundle"
"github.com/open-policy-agent/opa/download"
bundleUtils "github.com/open-policy-agent/opa/internal/bundle"
"github.com/open-policy-agent/opa/logging"
"github.com/open-policy-agent/opa/metrics"
"github.com/open-policy-agent/opa/plugins"
"github.com/open-policy-agent/opa/storage"
)
// maxActivationRetry represents the maximum number of attempts
// to activate persisted bundles. Activation retries are useful
// in scenarios where a persisted bundle may have a dependency on some
// other persisted bundle. As there are no ordering guarantees for which
// bundle loads first, retries could help in the bundle activation process.
// Typically, multiple bundles are not encouraged. The value chosen for
// maxActivationRetry allows upto 10 bundles to successfully activate
// in the worst case that they depend on each other. At the same time, it also
// ensures that too much time is not spent to activate bundles that will never
// successfully activate.
const maxActivationRetry = 10
// Loader defines the interface that the bundle plugin uses to control bundle
// loading via HTTP, disk, etc.
type Loader interface {
Start(context.Context)
Stop(context.Context)
Trigger(context.Context) error
SetCache(string)
ClearCache()
}
// Plugin implements bundle activation.
type Plugin struct {
config Config
manager *plugins.Manager // plugin manager for storage and service clients
status map[string]*Status // current status for each bundle
etags map[string]string // etag on last successful activation
listeners map[interface{}]func(Status) // listeners to send status updates to
bulkListeners map[interface{}]func(map[string]*Status) // listeners to send aggregated status updates to
downloaders map[string]Loader
logger logging.Logger
mtx sync.Mutex
cfgMtx sync.Mutex
ready bool
bundlePersistPath string
stopped bool
}
// New returns a new Plugin with the given config.
func New(parsedConfig *Config, manager *plugins.Manager) *Plugin {
initialStatus := map[string]*Status{}
for name := range parsedConfig.Bundles {
initialStatus[name] = &Status{
Name: name,
}
}
p := &Plugin{
manager: manager,
config: *parsedConfig,
status: initialStatus,
downloaders: make(map[string]Loader),
etags: make(map[string]string),
ready: false,
logger: manager.Logger(),
}
manager.UpdatePluginStatus(Name, &plugins.Status{State: plugins.StateNotReady})
return p
}
// Name identifies the plugin on manager.
const Name = "bundle"
// Lookup returns the bundle plugin registered with the manager.
func Lookup(manager *plugins.Manager) *Plugin {
if p := manager.Plugin(Name); p != nil {
return p.(*Plugin)
}
return nil
}
// Start runs the plugin. The plugin will periodically try to download bundles
// from the configured service. When a new bundle is downloaded, the data and
// policies are extracted and inserted into storage.
func (p *Plugin) Start(ctx context.Context) error {
p.mtx.Lock()
defer p.mtx.Unlock()
var err error
p.bundlePersistPath, err = p.getBundlePersistPath()
if err != nil {
return err
}
p.loadAndActivateBundlesFromDisk(ctx)
p.initDownloaders()
for name, dl := range p.downloaders {
p.log(name).Info("Starting bundle loader.")
dl.Start(ctx)
}
return nil
}
// Stop stops the plugin.
func (p *Plugin) Stop(ctx context.Context) {
p.mtx.Lock()
stopDownloaders := map[string]Loader{}
for name, dl := range p.downloaders {
stopDownloaders[name] = dl
}
p.downloaders = nil
p.stopped = true
p.mtx.Unlock()
for name, dl := range stopDownloaders {
p.log(name).Info("Stopping bundle loader.")
dl.Stop(ctx)
}
}
// Reconfigure notifies the plugin that it's configuration has changed.
// Any bundle configs that have changed or been added/removed will take
// affect.
func (p *Plugin) Reconfigure(ctx context.Context, config interface{}) {
// Reconfiguring should not occur in parallel, lock to ensure
// nothing swaps underneath us with the current p.config and the updated one.
// Use p.cfgMtx instead of p.mtx so as to not block any bundle downloads/activations
// that are in progress. We upgrade to p.mtx locking after stopping downloaders.
p.cfgMtx.Lock()
defer p.cfgMtx.Unlock()
// Look for any bundles that have had their config changed, are new, or have been removed
newConfig := config.(*Config)
newBundles, updatedBundles, deletedBundles := p.configDelta(newConfig)
p.config = *newConfig
if len(updatedBundles) == 0 && len(newBundles) == 0 && len(deletedBundles) == 0 {
// no relevant config changes
return
}
// Stop the downloaders outside p.mtx to allow them to finish handling any in-progress requests.
for name, dl := range p.downloaders {
_, updated := updatedBundles[name]
_, deleted := deletedBundles[name]
if updated || deleted {
dl.Stop(ctx)
}
}
// Only lock p.mtx once we start changing the internal maps
// and downloader configs.
p.mtx.Lock()
defer p.mtx.Unlock()
// Cleanup existing downloaders that are deleted
for name := range p.downloaders {
if _, deleted := deletedBundles[name]; deleted {
p.log(name).Info("Bundle loader configuration removed. Stopping bundle loader.")
delete(p.downloaders, name)
delete(p.status, name)
delete(p.etags, name)
}
}
// Deactivate the bundles that were removed
params := storage.WriteParams
params.Context = storage.NewContext() // TODO(sr): metrics?
err := storage.Txn(ctx, p.manager.Store, params, func(txn storage.Transaction) error {
opts := &bundle.DeactivateOpts{
Ctx: ctx,
Store: p.manager.Store,
Txn: txn,
BundleNames: deletedBundles,
}
err := bundle.Deactivate(opts)
if err != nil {
p.manager.Logger().Error(fmt.Sprint(deletedBundles), "Failed to deactivate bundles: %s", err)
return err
}
return nil
})
if err != nil {
// TODO(patrick-east): This probably shouldn't panic.. But OPA shouldn't
// continue in a potentially inconsistent state.
panic(errors.New("Unable deactivate bundle: " + err.Error()))
}
readyNow := p.ready
for name, source := range p.config.Bundles {
_, updated := updatedBundles[name]
_, isNew := newBundles[name]
if isNew || updated {
if isNew {
p.status[name] = &Status{Name: name}
p.log(name).Info("New bundle loader configuration added. Starting bundle loader.")
} else {
p.log(name).Info("Bundle loader configuration changed. Restarting bundle loader.")
}
p.downloaders[name] = p.newDownloader(name, source)
p.downloaders[name].Start(ctx)
readyNow = false
}
}
if !readyNow {
p.ready = false
p.manager.UpdatePluginStatus(Name, &plugins.Status{State: plugins.StateNotReady})
}
}
// Loaders returns the map of bundle loaders configured on this plugin.
func (p *Plugin) Loaders() map[string]Loader {
return p.downloaders
}
// Trigger triggers a bundle download on all configured bundles.
func (p *Plugin) Trigger(ctx context.Context) error {
p.mtx.Lock()
downloaders := map[string]Loader{}
for name, dl := range p.downloaders {
downloaders[name] = dl
}
p.mtx.Unlock()
for _, d := range downloaders {
// plugin callback will log the trigger error and include it in the bundle status
_ = d.Trigger(ctx)
}
return nil
}
// Register a listener to receive status updates. The name must be comparable.
// The listener will receive a status update for each bundle configured, they are
// not going to be aggregated. For all status updates use `RegisterBulkListener`.
func (p *Plugin) Register(name interface{}, listener func(Status)) {
p.mtx.Lock()
defer p.mtx.Unlock()
if p.listeners == nil {
p.listeners = map[interface{}]func(Status){}
}
p.listeners[name] = listener
}
// Unregister a listener to stop receiving status updates.
func (p *Plugin) Unregister(name interface{}) {
p.mtx.Lock()
defer p.mtx.Unlock()
delete(p.listeners, name)
}
// RegisterBulkListener registers a listener to receive bulk (aggregated) status updates. The name must be comparable.
func (p *Plugin) RegisterBulkListener(name interface{}, listener func(map[string]*Status)) {
p.mtx.Lock()
defer p.mtx.Unlock()
if p.bulkListeners == nil {
p.bulkListeners = map[interface{}]func(map[string]*Status){}
}
p.bulkListeners[name] = listener
}
// UnregisterBulkListener unregisters a listener to stop receiving aggregated status updates.
func (p *Plugin) UnregisterBulkListener(name interface{}) {
p.mtx.Lock()
defer p.mtx.Unlock()
delete(p.bulkListeners, name)
}
// Config returns the plugins current configuration
func (p *Plugin) Config() *Config {
return &p.config
}
func (p *Plugin) initDownloaders() {
// Initialize a downloader for each bundle configured.
for name, source := range p.config.Bundles {
p.downloaders[name] = p.newDownloader(name, source)
}
}
func (p *Plugin) loadAndActivateBundlesFromDisk(ctx context.Context) {
persistedBundles := map[string]*bundle.Bundle{}
for name, src := range p.config.Bundles {
if p.persistBundle(name) {
b, err := loadBundleFromDisk(p.bundlePersistPath, name, src)
if err != nil {
p.log(name).Error("Failed to load bundle from disk: %v", err)
p.status[name].SetError(err)
continue
}
if b == nil {
continue
}
persistedBundles[name] = b
}
}
if len(persistedBundles) == 0 {
return
}
for retry := 0; retry < maxActivationRetry; retry++ {
numActivatedBundles := 0
for name, b := range persistedBundles {
p.status[name].Metrics = metrics.New()
err := p.activate(ctx, name, b)
if err != nil {
p.log(name).Error("Bundle activation failed: %v", err)
p.status[name].SetError(err)
continue
}
p.status[name].SetError(nil)
p.status[name].SetActivateSuccess(b.Manifest.Revision)
p.checkPluginReadiness()
p.log(name).Debug("Bundle loaded from disk and activated successfully.")
numActivatedBundles++
}
if numActivatedBundles == len(persistedBundles) {
return
}
}
}
func (p *Plugin) newDownloader(name string, source *Source) Loader {
if u, err := url.Parse(source.Resource); err == nil {
switch u.Scheme {
case "file":
return &fileLoader{
name: name,
path: u.Path,
bvc: source.Signing,
sizeLimitBytes: source.SizeLimitBytes,
f: p.oneShot,
}
}
}
conf := source.Config
client := p.manager.Client(source.Service)
path := source.Resource
callback := func(ctx context.Context, u download.Update) {
// wrap the callback to include the name of the bundle that was updated
p.oneShot(ctx, name, u)
}
return download.New(conf, client, path).
WithCallback(callback).
WithBundleVerificationConfig(source.Signing).
WithSizeLimitBytes(source.SizeLimitBytes).
WithBundlePersistence(p.persistBundle(name))
}
func (p *Plugin) oneShot(ctx context.Context, name string, u download.Update) {
p.mtx.Lock()
defer p.mtx.Unlock()
p.process(ctx, name, u)
for _, listener := range p.listeners {
listener(*p.status[name])
}
for _, listener := range p.bulkListeners {
// Send a copy of the full status map to the bulk listeners.
// They shouldn't have access to the original underlying
// map, primarily for thread safety issues with modifications
// made to it.
statusCpy := map[string]*Status{}
for k, v := range p.status {
v := *v
statusCpy[k] = &v
}
listener(statusCpy)
}
}
func (p *Plugin) process(ctx context.Context, name string, u download.Update) {
if u.Metrics != nil {
p.status[name].Metrics = u.Metrics
} else {
p.status[name].Metrics = metrics.New()
}
p.status[name].SetRequest()
if u.Error != nil {
p.log(name).Error("Bundle load failed: %v", u.Error)
p.status[name].SetError(u.Error)
if !p.stopped {
etag := p.etags[name]
p.downloaders[name].SetCache(etag)
}
return
}
p.status[name].LastSuccessfulRequest = p.status[name].LastRequest
if u.Bundle != nil {
p.status[name].LastSuccessfulDownload = p.status[name].LastSuccessfulRequest
p.status[name].Metrics.Timer(metrics.RegoLoadBundles).Start()
defer p.status[name].Metrics.Timer(metrics.RegoLoadBundles).Stop()
if err := p.activate(ctx, name, u.Bundle); err != nil {
p.log(name).Error("Bundle activation failed: %v", err)
p.status[name].SetError(err)
if !p.stopped {
etag := p.etags[name]
p.downloaders[name].SetCache(etag)
}
return
}
if u.Bundle.Type() == bundle.SnapshotBundleType && p.persistBundle(name) {
p.log(name).Debug("Persisting bundle to disk in progress.")
err := p.saveBundleToDisk(name, u.Raw)
if err != nil {
p.log(name).Error("Persisting bundle to disk failed: %v", err)
p.status[name].SetError(err)
if !p.stopped {
etag := p.etags[name]
p.downloaders[name].SetCache(etag)
}
return
}
p.log(name).Debug("Bundle persisted to disk successfully at path %v.", filepath.Join(p.bundlePersistPath, name))
}
p.status[name].SetError(nil)
p.status[name].SetActivateSuccess(u.Bundle.Manifest.Revision)
if u.ETag != "" {
p.log(name).Info("Bundle loaded and activated successfully. Etag updated to %v.", u.ETag)
} else {
p.log(name).Info("Bundle loaded and activated successfully.")
}
p.etags[name] = u.ETag
// If the plugin wasn't ready yet then check if we are now after activating this bundle.
p.checkPluginReadiness()
return
}
if etag, ok := p.etags[name]; ok && u.ETag == etag {
p.log(name).Debug("Bundle load skipped, server replied with not modified.")
p.status[name].SetError(nil)
return
}
}
func (p *Plugin) checkPluginReadiness() {
if !p.ready {
readyNow := true // optimistically
for _, status := range p.status {
if len(status.Errors) > 0 || (status.LastSuccessfulActivation == time.Time{}) {
readyNow = false // Not ready yet, check again on next bundle activation.
break
}
}
if readyNow {
p.ready = true
p.manager.UpdatePluginStatus(Name, &plugins.Status{State: plugins.StateOK})
}
}
}
func (p *Plugin) activate(ctx context.Context, name string, b *bundle.Bundle) error {
p.log(name).Debug("Bundle activation in progress. Opening storage transaction.")
params := storage.WriteParams
params.Context = storage.NewContext().WithMetrics(p.status[name].Metrics)
err := storage.Txn(ctx, p.manager.Store, params, func(txn storage.Transaction) error {
p.log(name).Debug("Opened storage transaction (%v).", txn.ID())
defer p.log(name).Debug("Closing storage transaction (%v).", txn.ID())
// Compile the bundle modules with a new compiler and set it on the
// transaction params for use by onCommit hooks.
// If activating a delta bundle, use the manager's compiler which should have
// the polices compiled on it.
var compiler *ast.Compiler
if b.Type() == bundle.DeltaBundleType {
compiler = p.manager.GetCompiler()
}
if compiler == nil {
compiler = ast.NewCompiler()
}
compiler = compiler.WithPathConflictsCheck(storage.NonEmpty(ctx, p.manager.Store, txn)).
WithEnablePrintStatements(p.manager.EnablePrintStatements())
var activateErr error
opts := &bundle.ActivateOpts{
Ctx: ctx,
Store: p.manager.Store,
Txn: txn,
TxnCtx: params.Context,
Compiler: compiler,
Metrics: p.status[name].Metrics,
Bundles: map[string]*bundle.Bundle{name: b},
}
if p.config.IsMultiBundle() {
activateErr = bundle.Activate(opts)
} else {
activateErr = bundle.ActivateLegacy(opts)
}
plugins.SetCompilerOnContext(params.Context, compiler)
resolvers, err := bundleUtils.LoadWasmResolversFromStore(ctx, p.manager.Store, txn, nil)
if err != nil {
return err
}
plugins.SetWasmResolversOnContext(params.Context, resolvers)
return activateErr
})
return err
}
func (p *Plugin) persistBundle(name string) bool {
bundleSrc := p.config.Bundles[name]
if bundleSrc == nil {
return false
}
return bundleSrc.Persist
}
// configDelta will return a map of new bundle sources, updated bundle sources, and a set of deleted bundle names
func (p *Plugin) configDelta(newConfig *Config) (map[string]*Source, map[string]*Source, map[string]struct{}) {
deletedBundles := map[string]struct{}{}
for name := range p.config.Bundles {
deletedBundles[name] = struct{}{}
}
newBundles := map[string]*Source{}
updatedBundles := map[string]*Source{}
for name, source := range newConfig.Bundles {
oldSource, found := p.config.Bundles[name]
if !found {
newBundles[name] = source
} else {
delete(deletedBundles, name)
if !reflect.DeepEqual(oldSource, source) {
updatedBundles[name] = source
}
}
}
return newBundles, updatedBundles, deletedBundles
}
func (p *Plugin) saveBundleToDisk(name string, raw io.Reader) error {
bundleDir := filepath.Join(p.bundlePersistPath, name)
tmpFile := filepath.Join(bundleDir, ".bundle.tar.gz.tmp")
bundleFile := filepath.Join(bundleDir, "bundle.tar.gz")
saveErr := saveCurrentBundleToDisk(bundleDir, ".bundle.tar.gz.tmp", raw)
if saveErr != nil {
p.log(name).Error("Failed to save new bundle to disk: %v", saveErr)
if err := os.Remove(tmpFile); err != nil {
p.log(name).Warn("Failed to remove temp file ('%s'): %v", tmpFile, err)
}
if _, err := os.Stat(bundleFile); err == nil {
p.log(name).Warn("Older version of activated bundle persisted, ignoring error")
return nil
}
return saveErr
}
return os.Rename(tmpFile, bundleFile)
}
func saveCurrentBundleToDisk(path, filename string, raw io.Reader) error {
if _, err := os.Stat(path); os.IsNotExist(err) {
err = os.MkdirAll(path, os.ModePerm)
if err != nil {
return err
}
}
if raw == nil {
return fmt.Errorf("no raw bundle bytes to persist to disk")
}
dest, err := os.OpenFile(filepath.Join(path, filename), os.O_WRONLY|os.O_CREATE|os.O_TRUNC, 0644)
if err != nil {
return err
}
defer dest.Close()
_, err = io.Copy(dest, raw)
return err
}
func loadBundleFromDisk(path, name string, src *Source) (*bundle.Bundle, error) {
bundlePath := filepath.Join(path, name, "bundle.tar.gz")
if _, err := os.Stat(bundlePath); err == nil {
f, err := os.Open(filepath.Join(bundlePath))
if err != nil {
return nil, err
}
defer f.Close()
r := bundle.NewReader(f)
if src != nil {
r = r.WithBundleVerificationConfig(src.Signing)
}
b, err := r.Read()
if err != nil {
return nil, err
}
return &b, nil
} else if os.IsNotExist(err) {
return nil, nil
} else {
return nil, err
}
}
func (p *Plugin) log(name string) logging.Logger {
if p.logger == nil {
p.logger = logging.Get()
}
return p.logger.WithFields(map[string]interface{}{"name": name, "plugin": Name})
}
func (p *Plugin) getBundlePersistPath() (string, error) {
persistDir, err := p.manager.Config.GetPersistenceDirectory()
if err != nil {
return "", err
}
return filepath.Join(persistDir, "bundles"), nil
}
type fileLoader struct {
name string
path string
bvc *bundle.VerificationConfig
sizeLimitBytes int64
f func(context.Context, string, download.Update)
}
func (fl *fileLoader) Start(ctx context.Context) {
go func() {
fl.oneShot(ctx)
}()
}
func (*fileLoader) Stop(context.Context) {
}
func (*fileLoader) ClearCache() {
}
func (*fileLoader) SetCache(string) {
}
func (fl *fileLoader) Trigger(ctx context.Context) error {
fl.oneShot(ctx)
return nil
}
func (fl *fileLoader) oneShot(ctx context.Context) {
var u download.Update
u.Metrics = metrics.New()
info, err := os.Stat(fl.path)
u.Error = err
if err != nil {
fl.f(ctx, fl.name, u)
return
}
var reader *bundle.Reader
if info.IsDir() {
reader = bundle.NewCustomReader(bundle.NewDirectoryLoader(fl.path))
} else {
f, err := os.Open(fl.path)
u.Error = err
if err != nil {
fl.f(ctx, fl.name, u)
return
}
defer f.Close()
reader = bundle.NewReader(f)
}
b, err := reader.
WithMetrics(u.Metrics).
WithBundleVerificationConfig(fl.bvc).
WithSizeLimitBytes(fl.sizeLimitBytes).Read()
u.Error = err
if err == nil {
u.Bundle = &b
}
fl.f(ctx, fl.name, u)
}