repo_cache.go

  1package cache
  2
  3import (
  4	"fmt"
  5	"io"
  6	"os"
  7	"strconv"
  8	"sync"
  9
 10	"github.com/git-bug/git-bug/entities/board"
 11	"github.com/git-bug/git-bug/entities/bug"
 12	"github.com/git-bug/git-bug/entities/identity"
 13	"github.com/git-bug/git-bug/entity"
 14	"github.com/git-bug/git-bug/repository"
 15	"github.com/git-bug/git-bug/util/multierr"
 16	"github.com/git-bug/git-bug/util/process"
 17)
 18
 19// 1: original format
 20// 2: added cache for identities with a reference in the bug cache
 21// 3: no more legacy identity
 22// 4: entities make their IDs from data, not git commit
 23const formatVersion = 4
 24
 25// The maximum number of bugs loaded in memory. After that, eviction will be done.
 26const defaultMaxLoadedBugs = 1000
 27
 28var _ repository.RepoCommon = &RepoCache{}
 29var _ repository.RepoConfig = &RepoCache{}
 30var _ repository.RepoKeyring = &RepoCache{}
 31
 32// cacheMgmt is the expected interface for a sub-cache.
 33type cacheMgmt interface {
 34	Typename() string
 35	Load() error
 36	Build() <-chan BuildEvent
 37	SetCacheSize(size int)
 38	RemoveAll() error
 39	MergeAll(remote string) <-chan entity.MergeResult
 40	GetNamespace() string
 41	Close() error
 42}
 43
 44// RepoCache is a cache for a Repository. This cache has multiple functions:
 45//
 46//  1. After being loaded, a Bug is kept in memory in the cache, allowing for fast
 47//     access later.
 48//  2. The cache maintain in memory and on disk a pre-digested excerpt for each bug,
 49//     allowing for fast querying the whole set of bugs without having to load
 50//     them individually.
 51//  3. The cache guarantee that a single instance of a Bug is loaded at once, avoiding
 52//     loss of data that we could have with multiple copies in the same process.
 53//  4. The same way, the cache maintain in memory a single copy of the loaded identities.
 54//
 55// The cache also protect the on-disk data by locking the git repository for its
 56// own usage, by writing a lock file. Of course, normal git operations are not
 57// affected, only git-bug related one.
 58type RepoCache struct {
 59	// the underlying repo
 60	repo repository.ClockedRepo
 61
 62	// the name of the repository, as defined in the MultiRepoCache
 63	name string
 64
 65	// resolvers for all known entities and excerpts
 66	resolvers entity.Resolvers
 67
 68	boards     *RepoCacheBoard
 69	bugs       *RepoCacheBug
 70	identities *RepoCacheIdentity
 71
 72	subcaches []cacheMgmt
 73
 74	// the user identity's id, if known
 75	muUserIdentity sync.RWMutex
 76	userIdentityId entity.Id
 77}
 78
 79// NewRepoCache create or open a cache on top of a raw repository.
 80// The caller is expected to read all returned events before the cache is considered
 81// ready to use.
 82func NewRepoCache(r repository.ClockedRepo) (*RepoCache, chan BuildEvent) {
 83	return NewNamedRepoCache(r, defaultRepoName)
 84}
 85
 86// NewNamedRepoCache create or open a named cache on top of a raw repository.
 87// The caller is expected to read all returned events before the cache is considered
 88// ready to use.
 89func NewNamedRepoCache(r repository.ClockedRepo, name string) (*RepoCache, chan BuildEvent) {
 90	c := &RepoCache{
 91		repo: r,
 92		name: name,
 93	}
 94
 95	c.identities = NewRepoCacheIdentity(r, c.getResolvers, c.GetUserIdentity)
 96	c.subcaches = append(c.subcaches, c.identities)
 97
 98	c.bugs = NewRepoCacheBug(r, c.getResolvers, c.GetUserIdentity)
 99	c.subcaches = append(c.subcaches, c.bugs)
100
101	c.boards = NewRepoCacheBoard(r, c.getResolvers, c.GetUserIdentity)
102	c.subcaches = append(c.subcaches, c.boards)
103
104	c.resolvers = entity.Resolvers{
105		identity.Interface(nil): entity.ResolverFunc[*IdentityCache](c.identities.Resolve),
106		&IdentityCache{}:        entity.ResolverFunc[*IdentityCache](c.identities.Resolve),
107		&IdentityExcerpt{}:      entity.ResolverFunc[*IdentityExcerpt](c.identities.ResolveExcerpt),
108		bug.Interface(nil):      entity.ResolverFunc[*BugCache](c.bugs.Resolve),
109		&bug.Bug{}:              entity.ResolverFunc[*BugCache](c.bugs.Resolve),
110		&BugCache{}:             entity.ResolverFunc[*BugCache](c.bugs.Resolve),
111		&BugExcerpt{}:           entity.ResolverFunc[*BugExcerpt](c.bugs.ResolveExcerpt),
112		board.Interface(nil):    entity.ResolverFunc[*BoardCache](c.boards.Resolve),
113		&bug.Bug{}:              entity.ResolverFunc[*BoardCache](c.boards.Resolve),
114		&BoardCache{}:           entity.ResolverFunc[*BoardCache](c.boards.Resolve),
115		&BoardExcerpt{}:         entity.ResolverFunc[*BoardExcerpt](c.boards.ResolveExcerpt),
116	}
117
118	// small buffer so that below functions can emit an event without blocking
119	events := make(chan BuildEvent)
120
121	go func() {
122		defer close(events)
123
124		err := c.lock(events)
125		if err != nil {
126			events <- BuildEvent{Err: err}
127			return
128		}
129
130		err = c.load()
131		if err == nil {
132			return
133		}
134
135		// Cache is either missing, broken or outdated. Rebuilding.
136		c.buildCache(events)
137	}()
138
139	return c, events
140}
141
142func NewRepoCacheNoEvents(r repository.ClockedRepo) (*RepoCache, error) {
143	cache, events := NewRepoCache(r)
144	for event := range events {
145		if event.Err != nil {
146			for range events {
147			}
148			return nil, event.Err
149		}
150	}
151	return cache, nil
152}
153
154// Boards gives access to the Board entities
155func (c *RepoCache) Boards() *RepoCacheBoard {
156	return c.boards
157}
158
159// Bugs gives access to the Bug entities
160func (c *RepoCache) Bugs() *RepoCacheBug {
161	return c.bugs
162}
163
164// Identities gives access to the Identity entities
165func (c *RepoCache) Identities() *RepoCacheIdentity {
166	return c.identities
167}
168
169func (c *RepoCache) getResolvers() entity.Resolvers {
170	return c.resolvers
171}
172
173// setCacheSize change the maximum number of loaded bugs
174func (c *RepoCache) setCacheSize(size int) {
175	for _, subcache := range c.subcaches {
176		subcache.SetCacheSize(size)
177	}
178}
179
180// load will try to read from the disk all the cache files
181func (c *RepoCache) load() error {
182	var errWait multierr.ErrWaitGroup
183	for _, mgmt := range c.subcaches {
184		errWait.Go(mgmt.Load)
185	}
186	return errWait.Wait()
187}
188
189func (c *RepoCache) lock(events chan BuildEvent) error {
190	err := repoIsAvailable(c.repo, events)
191	if err != nil {
192		return err
193	}
194
195	f, err := c.repo.LocalStorage().Create(lockfile)
196	if err != nil {
197		return err
198	}
199
200	pid := fmt.Sprintf("%d", os.Getpid())
201	_, err = f.Write([]byte(pid))
202	if err != nil {
203		_ = f.Close()
204		return err
205	}
206
207	return f.Close()
208}
209
210func (c *RepoCache) Close() error {
211	var errWait multierr.ErrWaitGroup
212	for _, mgmt := range c.subcaches {
213		errWait.Go(mgmt.Close)
214	}
215	err := errWait.Wait()
216	if err != nil {
217		return err
218	}
219
220	err = c.repo.Close()
221	if err != nil {
222		return err
223	}
224
225	return c.repo.LocalStorage().Remove(lockfile)
226}
227
228type BuildEventType int
229
230const (
231	_ BuildEventType = iota
232	// BuildEventCacheIsBuilt signal that the cache is being built (aka, not skipped)
233	BuildEventCacheIsBuilt
234	// BuildEventRemoveLock signal that an old repo lock has been cleaned
235	BuildEventRemoveLock
236	// BuildEventStarted signal the beginning of a cache build for an entity
237	BuildEventStarted
238	// BuildEventProgress signal progress in the cache building for an entity
239	BuildEventProgress
240	// BuildEventFinished signal the end of a cache build for an entity
241	BuildEventFinished
242)
243
244// BuildEvent carry an event happening during the cache build process.
245type BuildEvent struct {
246	// Err carry an error if the build process failed. If set, no other field matter.
247	Err error
248	// Typename is the name of the entity of which the event relate to. Can be empty if not particular entity is involved.
249	Typename string
250	// Event is the type of the event.
251	Event BuildEventType
252	// Total is the total number of element being built. Set if Event is BuildEventStarted.
253	Total int64
254	// Progress is the current count of processed element. Set if Event is BuildEventProgress.
255	Progress int64
256}
257
258func (c *RepoCache) buildCache(events chan BuildEvent) {
259	events <- BuildEvent{Event: BuildEventCacheIsBuilt}
260
261	var wg sync.WaitGroup
262	for _, subcache := range c.subcaches {
263		wg.Add(1)
264		go func(subcache cacheMgmt) {
265			defer wg.Done()
266
267			buildEvents := subcache.Build()
268			for buildEvent := range buildEvents {
269				events <- buildEvent
270				if buildEvent.Err != nil {
271					return
272				}
273			}
274		}(subcache)
275	}
276	wg.Wait()
277}
278
279// repoIsAvailable check is the given repository is locked by a Cache.
280// Note: this is a smart function that will clean the lock file if the
281// corresponding process is not there anymore.
282// If no error is returned, the repo is free to edit.
283func repoIsAvailable(repo repository.RepoStorage, events chan BuildEvent) error {
284	// Todo: this leave way for a racey access to the repo between the test
285	// if the file exist and the actual write. It's probably not a problem in
286	// practice because using a repository will be done from user interaction
287	// or in a context where a single instance of git-bug is already guaranteed
288	// (say, a server with the web UI running). But still, that might be nice to
289	// have a mutex or something to guard that.
290
291	// Todo: this will fail if somehow the filesystem is shared with another
292	// computer. Should add a configuration that prevent the cleaning of the
293	// lock file
294
295	f, err := repo.LocalStorage().Open(lockfile)
296	if err != nil && !os.IsNotExist(err) {
297		return err
298	}
299
300	if err == nil {
301		// lock file already exist
302		buf, err := io.ReadAll(io.LimitReader(f, 10))
303		if err != nil {
304			_ = f.Close()
305			return err
306		}
307
308		err = f.Close()
309		if err != nil {
310			return err
311		}
312
313		if len(buf) >= 10 {
314			return fmt.Errorf("the lock file should be < 10 bytes")
315		}
316
317		pid, err := strconv.Atoi(string(buf))
318		if err != nil {
319			return err
320		}
321
322		if process.IsRunning(pid) {
323			return fmt.Errorf("the repository you want to access is already locked by the process pid %d", pid)
324		}
325
326		// The lock file is just laying there after a crash, clean it
327
328		events <- BuildEvent{Event: BuildEventRemoveLock}
329
330		err = repo.LocalStorage().Remove(lockfile)
331		if err != nil {
332			return err
333		}
334	}
335
336	return nil
337}