1 // Copyright (C) The Arvados Authors. All rights reserved.
3 // SPDX-License-Identifier: AGPL-3.0
20 "git.arvados.org/arvados.git/lib/config"
21 "git.arvados.org/arvados.git/sdk/go/arvados"
22 "git.arvados.org/arvados.git/sdk/go/arvadostest"
23 "git.arvados.org/arvados.git/sdk/go/ctxlog"
24 "github.com/jmoiron/sqlx"
25 "github.com/prometheus/client_golang/prometheus"
26 "github.com/prometheus/common/expfmt"
27 check "gopkg.in/check.v1"
30 var _ = check.Suite(&runSuite{})
32 type reqTracker struct {
37 func (rt *reqTracker) Count() int {
43 func (rt *reqTracker) Add(req *http.Request) int {
46 rt.reqs = append(rt.reqs, *req)
50 var stubServices = []arvados.KeepService{
52 UUID: "zzzzz-bi6l4-000000000000000",
53 ServiceHost: "keep0.zzzzz.arvadosapi.com",
55 ServiceSSLFlag: false,
59 UUID: "zzzzz-bi6l4-000000000000001",
60 ServiceHost: "keep1.zzzzz.arvadosapi.com",
62 ServiceSSLFlag: false,
66 UUID: "zzzzz-bi6l4-000000000000002",
67 ServiceHost: "keep2.zzzzz.arvadosapi.com",
69 ServiceSSLFlag: false,
73 UUID: "zzzzz-bi6l4-000000000000003",
74 ServiceHost: "keep3.zzzzz.arvadosapi.com",
76 ServiceSSLFlag: false,
80 UUID: "zzzzz-bi6l4-h0a0xwut9qa6g3a",
81 ServiceHost: "keep.zzzzz.arvadosapi.com",
88 var stubMounts = map[string][]arvados.KeepMount{
89 "keep0.zzzzz.arvadosapi.com:25107": {{
90 UUID: "zzzzz-ivpuk-000000000000000",
91 DeviceID: "keep0-vol0",
93 "keep1.zzzzz.arvadosapi.com:25107": {{
94 UUID: "zzzzz-ivpuk-100000000000000",
95 DeviceID: "keep1-vol0",
97 "keep2.zzzzz.arvadosapi.com:25107": {{
98 UUID: "zzzzz-ivpuk-200000000000000",
99 DeviceID: "keep2-vol0",
101 "keep3.zzzzz.arvadosapi.com:25107": {{
102 UUID: "zzzzz-ivpuk-300000000000000",
103 DeviceID: "keep3-vol0",
107 // stubServer is an HTTP transport that intercepts and processes all
108 // requests using its own handlers.
109 type stubServer struct {
114 logf func(string, ...interface{})
117 // Start initializes the stub server and returns an *http.Client that
118 // uses the stub server to handle all requests.
120 // A stubServer that has been started should eventually be shut down
122 func (s *stubServer) Start() *http.Client {
123 // Set up a config.Client that forwards all requests to s.mux
124 // via s.srv. Test cases will attach handlers to s.mux to get
125 // the desired responses.
126 s.mux = http.NewServeMux()
127 s.srv = httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
131 w.Header().Set("Content-Type", "application/json")
132 s.mux.ServeHTTP(w, r)
134 return &http.Client{Transport: s}
137 func (s *stubServer) RoundTrip(req *http.Request) (*http.Response, error) {
138 w := httptest.NewRecorder()
139 s.mux.ServeHTTP(w, req)
140 return &http.Response{
142 Status: fmt.Sprintf("%d %s", w.Code, http.StatusText(w.Code)),
144 Body: ioutil.NopCloser(w.Body)}, nil
147 // Close releases resources used by the server.
148 func (s *stubServer) Close() {
152 func (s *stubServer) serveStatic(path, data string) *reqTracker {
154 s.mux.HandleFunc(path, func(w http.ResponseWriter, r *http.Request) {
157 ioutil.ReadAll(r.Body)
160 io.WriteString(w, data)
165 func (s *stubServer) serveCurrentUserAdmin() *reqTracker {
166 return s.serveStatic("/arvados/v1/users/current",
167 `{"uuid":"zzzzz-tpzed-000000000000000","is_admin":true,"is_active":true}`)
170 func (s *stubServer) serveCurrentUserNotAdmin() *reqTracker {
171 return s.serveStatic("/arvados/v1/users/current",
172 `{"uuid":"zzzzz-tpzed-000000000000000","is_admin":false,"is_active":true}`)
175 func (s *stubServer) serveDiscoveryDoc() *reqTracker {
176 return s.serveStatic("/discovery/v1/apis/arvados/v1/rest",
177 `{"defaultCollectionReplication":2}`)
180 func (s *stubServer) serveZeroCollections() *reqTracker {
181 return s.serveStatic("/arvados/v1/collections",
182 `{"items":[],"items_available":0}`)
185 func (s *stubServer) serveFooBarFileCollections() *reqTracker {
187 s.mux.HandleFunc("/arvados/v1/collections", func(w http.ResponseWriter, r *http.Request) {
190 if strings.Contains(r.Form.Get("filters"), `modified_at`) {
191 io.WriteString(w, `{"items_available":0,"items":[]}`)
193 io.WriteString(w, `{"items_available":3,"items":[
194 {"uuid":"zzzzz-4zz18-aaaaaaaaaaaaaaa","portable_data_hash":"fa7aeb5140e2848d39b416daeef4ffc5+45","manifest_text":". 37b51d194a7513e45b56f6524f2d51f2+3 0:3:bar\n","modified_at":"2014-02-03T17:22:54Z"},
195 {"uuid":"zzzzz-4zz18-ehbhgtheo8909or","portable_data_hash":"fa7aeb5140e2848d39b416daeef4ffc5+45","manifest_text":". 37b51d194a7513e45b56f6524f2d51f2+3 0:3:bar\n","modified_at":"2014-02-03T17:22:54Z"},
196 {"uuid":"zzzzz-4zz18-znfnqtbbv4spc3w","portable_data_hash":"1f4b0bc7583c2a7f9102c395f4ffc5e3+45","manifest_text":". acbd18db4cc2f85cedef654fccc4a4d8+3 0:3:foo\n","modified_at":"2014-02-03T17:22:54Z"}]}`)
202 func (s *stubServer) serveCollectionsButSkipOne() *reqTracker {
204 s.mux.HandleFunc("/arvados/v1/collections", func(w http.ResponseWriter, r *http.Request) {
207 if strings.Contains(r.Form.Get("filters"), `"modified_at","\u003c="`) {
208 io.WriteString(w, `{"items_available":3,"items":[]}`)
209 } else if strings.Contains(r.Form.Get("filters"), `"modified_at","\u003e`) {
210 io.WriteString(w, `{"items_available":0,"items":[]}`)
211 } else if strings.Contains(r.Form.Get("filters"), `"modified_at","="`) && strings.Contains(r.Form.Get("filters"), `"uuid","\u003e"`) {
212 io.WriteString(w, `{"items_available":0,"items":[]}`)
213 } else if strings.Contains(r.Form.Get("filters"), `"modified_at","=",null`) {
214 io.WriteString(w, `{"items_available":0,"items":[]}`)
216 io.WriteString(w, `{"items_available":2,"items":[
217 {"uuid":"zzzzz-4zz18-ehbhgtheo8909or","portable_data_hash":"fa7aeb5140e2848d39b416daeef4ffc5+45","manifest_text":". 37b51d194a7513e45b56f6524f2d51f2+3 0:3:bar\n","modified_at":"2014-02-03T17:22:54Z"},
218 {"uuid":"zzzzz-4zz18-znfnqtbbv4spc3w","portable_data_hash":"1f4b0bc7583c2a7f9102c395f4ffc5e3+45","manifest_text":". acbd18db4cc2f85cedef654fccc4a4d8+3 0:3:foo\n","modified_at":"2014-02-03T17:22:54Z"}]}`)
224 func (s *stubServer) serveZeroKeepServices() *reqTracker {
225 return s.serveJSON("/arvados/v1/keep_services", arvados.KeepServiceList{})
228 func (s *stubServer) serveKeepServices(svcs []arvados.KeepService) *reqTracker {
229 return s.serveJSON("/arvados/v1/keep_services", arvados.KeepServiceList{
230 ItemsAvailable: len(svcs),
235 func (s *stubServer) serveJSON(path string, resp interface{}) *reqTracker {
237 s.mux.HandleFunc(path, func(w http.ResponseWriter, r *http.Request) {
239 json.NewEncoder(w).Encode(resp)
244 func (s *stubServer) serveKeepstoreMounts() *reqTracker {
246 s.mux.HandleFunc("/mounts", func(w http.ResponseWriter, r *http.Request) {
248 json.NewEncoder(w).Encode(stubMounts[r.Host])
253 func (s *stubServer) serveKeepstoreIndexFoo4Bar1() *reqTracker {
255 s.mux.HandleFunc("/index/", func(w http.ResponseWriter, r *http.Request) {
257 if r.Host == "keep0.zzzzz.arvadosapi.com:25107" {
258 io.WriteString(w, "37b51d194a7513e45b56f6524f2d51f2+3 12345678\n")
260 fmt.Fprintf(w, "acbd18db4cc2f85cedef654fccc4a4d8+3 %d\n\n", 12345678+count)
262 for _, mounts := range stubMounts {
263 for i, mnt := range mounts {
265 s.mux.HandleFunc(fmt.Sprintf("/mounts/%s/blocks", mnt.UUID), func(w http.ResponseWriter, r *http.Request) {
267 if i == 0 && r.Host == "keep0.zzzzz.arvadosapi.com:25107" {
268 io.WriteString(w, "37b51d194a7513e45b56f6524f2d51f2+3 12345678\n")
271 fmt.Fprintf(w, "acbd18db4cc2f85cedef654fccc4a4d8+3 %d\n", 12345678+count)
280 func (s *stubServer) serveKeepstoreIndexFoo1() *reqTracker {
282 s.mux.HandleFunc("/index/", func(w http.ResponseWriter, r *http.Request) {
284 io.WriteString(w, "acbd18db4cc2f85cedef654fccc4a4d8+3 12345678\n\n")
286 for _, mounts := range stubMounts {
287 for i, mnt := range mounts {
289 s.mux.HandleFunc(fmt.Sprintf("/mounts/%s/blocks", mnt.UUID), func(w http.ResponseWriter, r *http.Request) {
292 io.WriteString(w, "acbd18db4cc2f85cedef654fccc4a4d8+3 12345678\n\n")
294 io.WriteString(w, "\n")
302 func (s *stubServer) serveKeepstoreTrash() *reqTracker {
303 return s.serveStatic("/trash", `{}`)
306 func (s *stubServer) serveKeepstorePull() *reqTracker {
307 return s.serveStatic("/pull", `{}`)
310 type runSuite struct {
312 config *arvados.Cluster
314 client *arvados.Client
317 func (s *runSuite) newServer(options *RunOptions) *Server {
321 RunOptions: *options,
322 Metrics: newMetrics(prometheus.NewRegistry()),
323 Logger: options.Logger,
324 Dumper: options.Dumper,
330 func (s *runSuite) SetUpTest(c *check.C) {
331 cfg, err := config.NewLoader(nil, ctxlog.TestLogger(c)).Load()
332 c.Assert(err, check.Equals, nil)
333 s.config, err = cfg.GetCluster("")
334 c.Assert(err, check.Equals, nil)
335 s.db, err = sqlx.Open("postgres", s.config.PostgreSQL.Connection.String())
336 c.Assert(err, check.IsNil)
338 s.config.Collections.BalancePeriod = arvados.Duration(time.Second)
339 arvadostest.SetServiceURL(&s.config.Services.Keepbalance, "http://localhost:/")
341 s.client = &arvados.Client{
343 APIHost: "zzzzz.arvadosapi.com",
344 Client: s.stub.Start()}
346 s.stub.serveDiscoveryDoc()
350 func (s *runSuite) TearDownTest(c *check.C) {
354 func (s *runSuite) TestRefuseZeroCollections(c *check.C) {
355 defer arvados.NewClientFromEnv().RequestAndDecode(nil, "POST", "database/reset", nil, nil)
356 _, err := s.db.Exec(`delete from collections`)
357 c.Assert(err, check.IsNil)
361 Logger: ctxlog.TestLogger(c),
363 s.stub.serveCurrentUserAdmin()
364 s.stub.serveZeroCollections()
365 s.stub.serveKeepServices(stubServices)
366 s.stub.serveKeepstoreMounts()
367 s.stub.serveKeepstoreIndexFoo4Bar1()
368 trashReqs := s.stub.serveKeepstoreTrash()
369 pullReqs := s.stub.serveKeepstorePull()
370 srv := s.newServer(&opts)
371 _, err = srv.runOnce()
372 c.Check(err, check.ErrorMatches, "received zero collections")
373 c.Check(trashReqs.Count(), check.Equals, 4)
374 c.Check(pullReqs.Count(), check.Equals, 0)
377 func (s *runSuite) TestRefuseNonAdmin(c *check.C) {
381 Logger: ctxlog.TestLogger(c),
383 s.stub.serveCurrentUserNotAdmin()
384 s.stub.serveZeroCollections()
385 s.stub.serveKeepServices(stubServices)
386 s.stub.serveKeepstoreMounts()
387 trashReqs := s.stub.serveKeepstoreTrash()
388 pullReqs := s.stub.serveKeepstorePull()
389 srv := s.newServer(&opts)
390 _, err := srv.runOnce()
391 c.Check(err, check.ErrorMatches, "current user .* is not .* admin user")
392 c.Check(trashReqs.Count(), check.Equals, 0)
393 c.Check(pullReqs.Count(), check.Equals, 0)
396 func (s *runSuite) TestWriteLostBlocks(c *check.C) {
397 lostf, err := ioutil.TempFile("", "keep-balance-lost-blocks-test-")
398 c.Assert(err, check.IsNil)
399 s.config.Collections.BlobMissingReport = lostf.Name()
400 defer os.Remove(lostf.Name())
404 Logger: ctxlog.TestLogger(c),
406 s.stub.serveCurrentUserAdmin()
407 s.stub.serveFooBarFileCollections()
408 s.stub.serveKeepServices(stubServices)
409 s.stub.serveKeepstoreMounts()
410 s.stub.serveKeepstoreIndexFoo1()
411 s.stub.serveKeepstoreTrash()
412 s.stub.serveKeepstorePull()
413 srv := s.newServer(&opts)
414 c.Assert(err, check.IsNil)
415 _, err = srv.runOnce()
416 c.Check(err, check.IsNil)
417 lost, err := ioutil.ReadFile(lostf.Name())
418 c.Assert(err, check.IsNil)
419 c.Check(string(lost), check.Matches, `(?ms).*37b51d194a7513e45b56f6524f2d51f2.* fa7aeb5140e2848d39b416daeef4ffc5\+45.*`)
422 func (s *runSuite) TestDryRun(c *check.C) {
426 Logger: ctxlog.TestLogger(c),
428 s.stub.serveCurrentUserAdmin()
429 collReqs := s.stub.serveFooBarFileCollections()
430 s.stub.serveKeepServices(stubServices)
431 s.stub.serveKeepstoreMounts()
432 s.stub.serveKeepstoreIndexFoo4Bar1()
433 trashReqs := s.stub.serveKeepstoreTrash()
434 pullReqs := s.stub.serveKeepstorePull()
435 srv := s.newServer(&opts)
436 bal, err := srv.runOnce()
437 c.Check(err, check.IsNil)
438 for _, req := range collReqs.reqs {
439 c.Check(req.Form.Get("include_trash"), check.Equals, "true")
440 c.Check(req.Form.Get("include_old_versions"), check.Equals, "true")
442 c.Check(trashReqs.Count(), check.Equals, 0)
443 c.Check(pullReqs.Count(), check.Equals, 0)
444 c.Check(bal.stats.pulls, check.Not(check.Equals), 0)
445 c.Check(bal.stats.underrep.replicas, check.Not(check.Equals), 0)
446 c.Check(bal.stats.overrep.replicas, check.Not(check.Equals), 0)
449 func (s *runSuite) TestCommit(c *check.C) {
450 s.config.Collections.BlobMissingReport = c.MkDir() + "/keep-balance-lost-blocks-test-"
451 s.config.ManagementToken = "xyzzy"
455 Logger: ctxlog.TestLogger(c),
456 Dumper: ctxlog.TestLogger(c),
458 s.stub.serveCurrentUserAdmin()
459 s.stub.serveFooBarFileCollections()
460 s.stub.serveKeepServices(stubServices)
461 s.stub.serveKeepstoreMounts()
462 s.stub.serveKeepstoreIndexFoo4Bar1()
463 trashReqs := s.stub.serveKeepstoreTrash()
464 pullReqs := s.stub.serveKeepstorePull()
465 srv := s.newServer(&opts)
466 bal, err := srv.runOnce()
467 c.Check(err, check.IsNil)
468 c.Check(trashReqs.Count(), check.Equals, 8)
469 c.Check(pullReqs.Count(), check.Equals, 4)
470 // "foo" block is overreplicated by 2
471 c.Check(bal.stats.trashes, check.Equals, 2)
472 // "bar" block is underreplicated by 1, and its only copy is
473 // in a poor rendezvous position
474 c.Check(bal.stats.pulls, check.Equals, 2)
476 lost, err := ioutil.ReadFile(s.config.Collections.BlobMissingReport)
477 c.Assert(err, check.IsNil)
478 c.Check(string(lost), check.Not(check.Matches), `(?ms).*acbd18db4cc2f85cedef654fccc4a4d8.*`)
480 buf, err := s.getMetrics(c, srv)
481 c.Check(err, check.IsNil)
482 bufstr := buf.String()
483 c.Check(bufstr, check.Matches, `(?ms).*\narvados_keep_total_bytes 15\n.*`)
484 c.Check(bufstr, check.Matches, `(?ms).*\narvados_keepbalance_changeset_compute_seconds_sum [0-9\.]+\n.*`)
485 c.Check(bufstr, check.Matches, `(?ms).*\narvados_keepbalance_changeset_compute_seconds_count 1\n.*`)
486 c.Check(bufstr, check.Matches, `(?ms).*\narvados_keep_dedup_byte_ratio [1-9].*`)
487 c.Check(bufstr, check.Matches, `(?ms).*\narvados_keep_dedup_block_ratio [1-9].*`)
490 func (s *runSuite) TestRunForever(c *check.C) {
491 s.config.ManagementToken = "xyzzy"
495 Logger: ctxlog.TestLogger(c),
496 Dumper: ctxlog.TestLogger(c),
498 s.stub.serveCurrentUserAdmin()
499 s.stub.serveFooBarFileCollections()
500 s.stub.serveKeepServices(stubServices)
501 s.stub.serveKeepstoreMounts()
502 s.stub.serveKeepstoreIndexFoo4Bar1()
503 trashReqs := s.stub.serveKeepstoreTrash()
504 pullReqs := s.stub.serveKeepstorePull()
506 stop := make(chan interface{})
507 s.config.Collections.BalancePeriod = arvados.Duration(time.Millisecond)
508 srv := s.newServer(&opts)
510 done := make(chan bool)
516 // Each run should send 4 pull lists + 4 trash lists. The
517 // first run should also send 4 empty trash lists at
518 // startup. We should complete all four runs in much less than
520 for t0 := time.Now(); pullReqs.Count() < 16 && time.Since(t0) < 10*time.Second; {
521 time.Sleep(time.Millisecond)
525 c.Check(pullReqs.Count() >= 16, check.Equals, true)
526 c.Check(trashReqs.Count(), check.Equals, pullReqs.Count()+4)
528 buf, err := s.getMetrics(c, srv)
529 c.Check(err, check.IsNil)
530 c.Check(buf, check.Matches, `(?ms).*\narvados_keepbalance_changeset_compute_seconds_count `+fmt.Sprintf("%d", pullReqs.Count()/4)+`\n.*`)
533 func (s *runSuite) getMetrics(c *check.C, srv *Server) (*bytes.Buffer, error) {
534 mfs, err := srv.Metrics.reg.Gather()
540 for _, mf := range mfs {
541 if _, err := expfmt.MetricFamilyToText(&buf, mf); err != nil {