Merge branch '14670-new-java-sdk-docs' refs #14670
[arvados.git] / services / keep-balance / balance_run_test.go
1 // Copyright (C) The Arvados Authors. All rights reserved.
2 //
3 // SPDX-License-Identifier: AGPL-3.0
4
5 package main
6
7 import (
8         "encoding/json"
9         "fmt"
10         "io"
11         "io/ioutil"
12         "net/http"
13         "net/http/httptest"
14         "strings"
15         "sync"
16         "time"
17
18         "git.curoverse.com/arvados.git/sdk/go/arvados"
19         "github.com/sirupsen/logrus"
20         check "gopkg.in/check.v1"
21 )
22
23 var _ = check.Suite(&runSuite{})
24
25 type reqTracker struct {
26         reqs []http.Request
27         sync.Mutex
28 }
29
30 func (rt *reqTracker) Count() int {
31         rt.Lock()
32         defer rt.Unlock()
33         return len(rt.reqs)
34 }
35
36 func (rt *reqTracker) Add(req *http.Request) int {
37         rt.Lock()
38         defer rt.Unlock()
39         rt.reqs = append(rt.reqs, *req)
40         return len(rt.reqs)
41 }
42
43 var stubServices = []arvados.KeepService{
44         {
45                 UUID:           "zzzzz-bi6l4-000000000000000",
46                 ServiceHost:    "keep0.zzzzz.arvadosapi.com",
47                 ServicePort:    25107,
48                 ServiceSSLFlag: false,
49                 ServiceType:    "disk",
50         },
51         {
52                 UUID:           "zzzzz-bi6l4-000000000000001",
53                 ServiceHost:    "keep1.zzzzz.arvadosapi.com",
54                 ServicePort:    25107,
55                 ServiceSSLFlag: false,
56                 ServiceType:    "disk",
57         },
58         {
59                 UUID:           "zzzzz-bi6l4-000000000000002",
60                 ServiceHost:    "keep2.zzzzz.arvadosapi.com",
61                 ServicePort:    25107,
62                 ServiceSSLFlag: false,
63                 ServiceType:    "disk",
64         },
65         {
66                 UUID:           "zzzzz-bi6l4-000000000000003",
67                 ServiceHost:    "keep3.zzzzz.arvadosapi.com",
68                 ServicePort:    25107,
69                 ServiceSSLFlag: false,
70                 ServiceType:    "disk",
71         },
72         {
73                 UUID:           "zzzzz-bi6l4-h0a0xwut9qa6g3a",
74                 ServiceHost:    "keep.zzzzz.arvadosapi.com",
75                 ServicePort:    25333,
76                 ServiceSSLFlag: true,
77                 ServiceType:    "proxy",
78         },
79 }
80
81 var stubMounts = map[string][]arvados.KeepMount{
82         "keep0.zzzzz.arvadosapi.com:25107": {{
83                 UUID:     "zzzzz-ivpuk-000000000000000",
84                 DeviceID: "keep0-vol0",
85         }},
86         "keep1.zzzzz.arvadosapi.com:25107": {{
87                 UUID:     "zzzzz-ivpuk-100000000000000",
88                 DeviceID: "keep1-vol0",
89         }},
90         "keep2.zzzzz.arvadosapi.com:25107": {{
91                 UUID:     "zzzzz-ivpuk-200000000000000",
92                 DeviceID: "keep2-vol0",
93         }},
94         "keep3.zzzzz.arvadosapi.com:25107": {{
95                 UUID:     "zzzzz-ivpuk-300000000000000",
96                 DeviceID: "keep3-vol0",
97         }},
98 }
99
100 // stubServer is an HTTP transport that intercepts and processes all
101 // requests using its own handlers.
102 type stubServer struct {
103         mux      *http.ServeMux
104         srv      *httptest.Server
105         mutex    sync.Mutex
106         Requests reqTracker
107         logf     func(string, ...interface{})
108 }
109
110 // Start initializes the stub server and returns an *http.Client that
111 // uses the stub server to handle all requests.
112 //
113 // A stubServer that has been started should eventually be shut down
114 // with Close().
115 func (s *stubServer) Start() *http.Client {
116         // Set up a config.Client that forwards all requests to s.mux
117         // via s.srv. Test cases will attach handlers to s.mux to get
118         // the desired responses.
119         s.mux = http.NewServeMux()
120         s.srv = httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
121                 s.mutex.Lock()
122                 s.Requests.Add(r)
123                 s.mutex.Unlock()
124                 w.Header().Set("Content-Type", "application/json")
125                 s.mux.ServeHTTP(w, r)
126         }))
127         return &http.Client{Transport: s}
128 }
129
130 func (s *stubServer) RoundTrip(req *http.Request) (*http.Response, error) {
131         w := httptest.NewRecorder()
132         s.mux.ServeHTTP(w, req)
133         return &http.Response{
134                 StatusCode: w.Code,
135                 Status:     fmt.Sprintf("%d %s", w.Code, http.StatusText(w.Code)),
136                 Header:     w.HeaderMap,
137                 Body:       ioutil.NopCloser(w.Body)}, nil
138 }
139
140 // Close releases resources used by the server.
141 func (s *stubServer) Close() {
142         s.srv.Close()
143 }
144
145 func (s *stubServer) serveStatic(path, data string) *reqTracker {
146         rt := &reqTracker{}
147         s.mux.HandleFunc(path, func(w http.ResponseWriter, r *http.Request) {
148                 rt.Add(r)
149                 if r.Body != nil {
150                         ioutil.ReadAll(r.Body)
151                         r.Body.Close()
152                 }
153                 io.WriteString(w, data)
154         })
155         return rt
156 }
157
158 func (s *stubServer) serveCurrentUserAdmin() *reqTracker {
159         return s.serveStatic("/arvados/v1/users/current",
160                 `{"uuid":"zzzzz-tpzed-000000000000000","is_admin":true,"is_active":true}`)
161 }
162
163 func (s *stubServer) serveCurrentUserNotAdmin() *reqTracker {
164         return s.serveStatic("/arvados/v1/users/current",
165                 `{"uuid":"zzzzz-tpzed-000000000000000","is_admin":false,"is_active":true}`)
166 }
167
168 func (s *stubServer) serveDiscoveryDoc() *reqTracker {
169         return s.serveStatic("/discovery/v1/apis/arvados/v1/rest",
170                 `{"defaultCollectionReplication":2}`)
171 }
172
173 func (s *stubServer) serveZeroCollections() *reqTracker {
174         return s.serveStatic("/arvados/v1/collections",
175                 `{"items":[],"items_available":0}`)
176 }
177
178 func (s *stubServer) serveFooBarFileCollections() *reqTracker {
179         rt := &reqTracker{}
180         s.mux.HandleFunc("/arvados/v1/collections", func(w http.ResponseWriter, r *http.Request) {
181                 r.ParseForm()
182                 rt.Add(r)
183                 if strings.Contains(r.Form.Get("filters"), `modified_at`) {
184                         io.WriteString(w, `{"items_available":0,"items":[]}`)
185                 } else {
186                         io.WriteString(w, `{"items_available":3,"items":[
187                                 {"uuid":"zzzzz-4zz18-aaaaaaaaaaaaaaa","portable_data_hash":"fa7aeb5140e2848d39b416daeef4ffc5+45","manifest_text":". 37b51d194a7513e45b56f6524f2d51f2+3 0:3:bar\n","modified_at":"2014-02-03T17:22:54Z"},
188                                 {"uuid":"zzzzz-4zz18-ehbhgtheo8909or","portable_data_hash":"fa7aeb5140e2848d39b416daeef4ffc5+45","manifest_text":". 37b51d194a7513e45b56f6524f2d51f2+3 0:3:bar\n","modified_at":"2014-02-03T17:22:54Z"},
189                                 {"uuid":"zzzzz-4zz18-znfnqtbbv4spc3w","portable_data_hash":"1f4b0bc7583c2a7f9102c395f4ffc5e3+45","manifest_text":". acbd18db4cc2f85cedef654fccc4a4d8+3 0:3:foo\n","modified_at":"2014-02-03T17:22:54Z"}]}`)
190                 }
191         })
192         return rt
193 }
194
195 func (s *stubServer) serveCollectionsButSkipOne() *reqTracker {
196         rt := &reqTracker{}
197         s.mux.HandleFunc("/arvados/v1/collections", func(w http.ResponseWriter, r *http.Request) {
198                 r.ParseForm()
199                 rt.Add(r)
200                 if strings.Contains(r.Form.Get("filters"), `"modified_at","\u003c="`) {
201                         io.WriteString(w, `{"items_available":3,"items":[]}`)
202                 } else if strings.Contains(r.Form.Get("filters"), `"modified_at","\u003e`) {
203                         io.WriteString(w, `{"items_available":0,"items":[]}`)
204                 } else if strings.Contains(r.Form.Get("filters"), `"modified_at","="`) && strings.Contains(r.Form.Get("filters"), `"uuid","\u003e"`) {
205                         io.WriteString(w, `{"items_available":0,"items":[]}`)
206                 } else if strings.Contains(r.Form.Get("filters"), `"modified_at","=",null`) {
207                         io.WriteString(w, `{"items_available":0,"items":[]}`)
208                 } else {
209                         io.WriteString(w, `{"items_available":2,"items":[
210                                 {"uuid":"zzzzz-4zz18-ehbhgtheo8909or","portable_data_hash":"fa7aeb5140e2848d39b416daeef4ffc5+45","manifest_text":". 37b51d194a7513e45b56f6524f2d51f2+3 0:3:bar\n","modified_at":"2014-02-03T17:22:54Z"},
211                                 {"uuid":"zzzzz-4zz18-znfnqtbbv4spc3w","portable_data_hash":"1f4b0bc7583c2a7f9102c395f4ffc5e3+45","manifest_text":". acbd18db4cc2f85cedef654fccc4a4d8+3 0:3:foo\n","modified_at":"2014-02-03T17:22:54Z"}]}`)
212                 }
213         })
214         return rt
215 }
216
217 func (s *stubServer) serveZeroKeepServices() *reqTracker {
218         return s.serveJSON("/arvados/v1/keep_services", arvados.KeepServiceList{})
219 }
220
221 func (s *stubServer) serveKeepServices(svcs []arvados.KeepService) *reqTracker {
222         return s.serveJSON("/arvados/v1/keep_services", arvados.KeepServiceList{
223                 ItemsAvailable: len(svcs),
224                 Items:          svcs,
225         })
226 }
227
228 func (s *stubServer) serveJSON(path string, resp interface{}) *reqTracker {
229         rt := &reqTracker{}
230         s.mux.HandleFunc(path, func(w http.ResponseWriter, r *http.Request) {
231                 rt.Add(r)
232                 json.NewEncoder(w).Encode(resp)
233         })
234         return rt
235 }
236
237 func (s *stubServer) serveKeepstoreMounts() *reqTracker {
238         rt := &reqTracker{}
239         s.mux.HandleFunc("/mounts", func(w http.ResponseWriter, r *http.Request) {
240                 rt.Add(r)
241                 json.NewEncoder(w).Encode(stubMounts[r.Host])
242         })
243         return rt
244 }
245
246 func (s *stubServer) serveKeepstoreIndexFoo4Bar1() *reqTracker {
247         rt := &reqTracker{}
248         s.mux.HandleFunc("/index/", func(w http.ResponseWriter, r *http.Request) {
249                 count := rt.Add(r)
250                 if r.Host == "keep0.zzzzz.arvadosapi.com:25107" {
251                         io.WriteString(w, "37b51d194a7513e45b56f6524f2d51f2+3 12345678\n")
252                 }
253                 fmt.Fprintf(w, "acbd18db4cc2f85cedef654fccc4a4d8+3 %d\n\n", 12345678+count)
254         })
255         for _, mounts := range stubMounts {
256                 for i, mnt := range mounts {
257                         i := i
258                         s.mux.HandleFunc(fmt.Sprintf("/mounts/%s/blocks", mnt.UUID), func(w http.ResponseWriter, r *http.Request) {
259                                 count := rt.Add(r)
260                                 if i == 0 && r.Host == "keep0.zzzzz.arvadosapi.com:25107" {
261                                         io.WriteString(w, "37b51d194a7513e45b56f6524f2d51f2+3 12345678\n")
262                                 }
263                                 if i == 0 {
264                                         fmt.Fprintf(w, "acbd18db4cc2f85cedef654fccc4a4d8+3 %d\n", 12345678+count)
265                                 }
266                                 fmt.Fprintf(w, "\n")
267                         })
268                 }
269         }
270         return rt
271 }
272
273 func (s *stubServer) serveKeepstoreTrash() *reqTracker {
274         return s.serveStatic("/trash", `{}`)
275 }
276
277 func (s *stubServer) serveKeepstorePull() *reqTracker {
278         return s.serveStatic("/pull", `{}`)
279 }
280
281 type runSuite struct {
282         stub   stubServer
283         config Config
284 }
285
286 // make a log.Logger that writes to the current test's c.Log().
287 func (s *runSuite) logger(c *check.C) *logrus.Logger {
288         r, w := io.Pipe()
289         go func() {
290                 buf := make([]byte, 10000)
291                 for {
292                         n, err := r.Read(buf)
293                         if n > 0 {
294                                 if buf[n-1] == '\n' {
295                                         n--
296                                 }
297                                 c.Log(string(buf[:n]))
298                         }
299                         if err != nil {
300                                 break
301                         }
302                 }
303         }()
304         logger := logrus.New()
305         logger.Out = w
306         return logger
307 }
308
309 func (s *runSuite) SetUpTest(c *check.C) {
310         s.config = Config{
311                 Client: arvados.Client{
312                         AuthToken: "xyzzy",
313                         APIHost:   "zzzzz.arvadosapi.com",
314                         Client:    s.stub.Start()},
315                 KeepServiceTypes: []string{"disk"},
316                 RunPeriod:        arvados.Duration(time.Second),
317         }
318         s.stub.serveDiscoveryDoc()
319         s.stub.logf = c.Logf
320 }
321
322 func (s *runSuite) TearDownTest(c *check.C) {
323         s.stub.Close()
324 }
325
326 func (s *runSuite) TestRefuseZeroCollections(c *check.C) {
327         opts := RunOptions{
328                 CommitPulls: true,
329                 CommitTrash: true,
330                 Logger:      s.logger(c),
331         }
332         s.stub.serveCurrentUserAdmin()
333         s.stub.serveZeroCollections()
334         s.stub.serveKeepServices(stubServices)
335         s.stub.serveKeepstoreMounts()
336         s.stub.serveKeepstoreIndexFoo4Bar1()
337         trashReqs := s.stub.serveKeepstoreTrash()
338         pullReqs := s.stub.serveKeepstorePull()
339         srv, err := NewServer(s.config, opts)
340         c.Assert(err, check.IsNil)
341         _, err = srv.Run()
342         c.Check(err, check.ErrorMatches, "received zero collections")
343         c.Check(trashReqs.Count(), check.Equals, 4)
344         c.Check(pullReqs.Count(), check.Equals, 0)
345 }
346
347 func (s *runSuite) TestServiceTypes(c *check.C) {
348         opts := RunOptions{
349                 CommitPulls: true,
350                 CommitTrash: true,
351                 Logger:      s.logger(c),
352         }
353         s.config.KeepServiceTypes = []string{"unlisted-type"}
354         s.stub.serveCurrentUserAdmin()
355         s.stub.serveFooBarFileCollections()
356         s.stub.serveKeepServices(stubServices)
357         s.stub.serveKeepstoreMounts()
358         indexReqs := s.stub.serveKeepstoreIndexFoo4Bar1()
359         trashReqs := s.stub.serveKeepstoreTrash()
360         srv, err := NewServer(s.config, opts)
361         c.Assert(err, check.IsNil)
362         _, err = srv.Run()
363         c.Check(err, check.IsNil)
364         c.Check(indexReqs.Count(), check.Equals, 0)
365         c.Check(trashReqs.Count(), check.Equals, 0)
366 }
367
368 func (s *runSuite) TestRefuseNonAdmin(c *check.C) {
369         opts := RunOptions{
370                 CommitPulls: true,
371                 CommitTrash: true,
372                 Logger:      s.logger(c),
373         }
374         s.stub.serveCurrentUserNotAdmin()
375         s.stub.serveZeroCollections()
376         s.stub.serveKeepServices(stubServices)
377         s.stub.serveKeepstoreMounts()
378         trashReqs := s.stub.serveKeepstoreTrash()
379         pullReqs := s.stub.serveKeepstorePull()
380         srv, err := NewServer(s.config, opts)
381         c.Assert(err, check.IsNil)
382         _, err = srv.Run()
383         c.Check(err, check.ErrorMatches, "current user .* is not .* admin user")
384         c.Check(trashReqs.Count(), check.Equals, 0)
385         c.Check(pullReqs.Count(), check.Equals, 0)
386 }
387
388 func (s *runSuite) TestDetectSkippedCollections(c *check.C) {
389         opts := RunOptions{
390                 CommitPulls: true,
391                 CommitTrash: true,
392                 Logger:      s.logger(c),
393         }
394         s.stub.serveCurrentUserAdmin()
395         s.stub.serveCollectionsButSkipOne()
396         s.stub.serveKeepServices(stubServices)
397         s.stub.serveKeepstoreMounts()
398         s.stub.serveKeepstoreIndexFoo4Bar1()
399         trashReqs := s.stub.serveKeepstoreTrash()
400         pullReqs := s.stub.serveKeepstorePull()
401         srv, err := NewServer(s.config, opts)
402         c.Assert(err, check.IsNil)
403         _, err = srv.Run()
404         c.Check(err, check.ErrorMatches, `Retrieved 2 collections with modtime <= .* but server now reports there are 3 collections.*`)
405         c.Check(trashReqs.Count(), check.Equals, 4)
406         c.Check(pullReqs.Count(), check.Equals, 0)
407 }
408
409 func (s *runSuite) TestDryRun(c *check.C) {
410         opts := RunOptions{
411                 CommitPulls: false,
412                 CommitTrash: false,
413                 Logger:      s.logger(c),
414         }
415         s.stub.serveCurrentUserAdmin()
416         collReqs := s.stub.serveFooBarFileCollections()
417         s.stub.serveKeepServices(stubServices)
418         s.stub.serveKeepstoreMounts()
419         s.stub.serveKeepstoreIndexFoo4Bar1()
420         trashReqs := s.stub.serveKeepstoreTrash()
421         pullReqs := s.stub.serveKeepstorePull()
422         srv, err := NewServer(s.config, opts)
423         c.Assert(err, check.IsNil)
424         bal, err := srv.Run()
425         c.Check(err, check.IsNil)
426         for _, req := range collReqs.reqs {
427                 c.Check(req.Form.Get("include_trash"), check.Equals, "true")
428                 c.Check(req.Form.Get("include_old_versions"), check.Equals, "true")
429         }
430         c.Check(trashReqs.Count(), check.Equals, 0)
431         c.Check(pullReqs.Count(), check.Equals, 0)
432         c.Check(bal.stats.pulls, check.Not(check.Equals), 0)
433         c.Check(bal.stats.underrep.replicas, check.Not(check.Equals), 0)
434         c.Check(bal.stats.overrep.replicas, check.Not(check.Equals), 0)
435 }
436
437 func (s *runSuite) TestCommit(c *check.C) {
438         s.config.Listen = ":"
439         s.config.ManagementToken = "xyzzy"
440         opts := RunOptions{
441                 CommitPulls: true,
442                 CommitTrash: true,
443                 Logger:      s.logger(c),
444                 Dumper:      s.logger(c),
445         }
446         s.stub.serveCurrentUserAdmin()
447         s.stub.serveFooBarFileCollections()
448         s.stub.serveKeepServices(stubServices)
449         s.stub.serveKeepstoreMounts()
450         s.stub.serveKeepstoreIndexFoo4Bar1()
451         trashReqs := s.stub.serveKeepstoreTrash()
452         pullReqs := s.stub.serveKeepstorePull()
453         srv, err := NewServer(s.config, opts)
454         c.Assert(err, check.IsNil)
455         bal, err := srv.Run()
456         c.Check(err, check.IsNil)
457         c.Check(trashReqs.Count(), check.Equals, 8)
458         c.Check(pullReqs.Count(), check.Equals, 4)
459         // "foo" block is overreplicated by 2
460         c.Check(bal.stats.trashes, check.Equals, 2)
461         // "bar" block is underreplicated by 1, and its only copy is
462         // in a poor rendezvous position
463         c.Check(bal.stats.pulls, check.Equals, 2)
464
465         metrics := s.getMetrics(c, srv)
466         c.Check(metrics, check.Matches, `(?ms).*\narvados_keep_total_bytes 15\n.*`)
467         c.Check(metrics, check.Matches, `(?ms).*\narvados_keepbalance_changeset_compute_seconds_sum [0-9\.]+\n.*`)
468         c.Check(metrics, check.Matches, `(?ms).*\narvados_keepbalance_changeset_compute_seconds_count 1\n.*`)
469         c.Check(metrics, check.Matches, `(?ms).*\narvados_keep_dedup_byte_ratio 1\.5\n.*`)
470         c.Check(metrics, check.Matches, `(?ms).*\narvados_keep_dedup_block_ratio 1\.5\n.*`)
471 }
472
473 func (s *runSuite) TestRunForever(c *check.C) {
474         s.config.Listen = ":"
475         s.config.ManagementToken = "xyzzy"
476         opts := RunOptions{
477                 CommitPulls: true,
478                 CommitTrash: true,
479                 Logger:      s.logger(c),
480                 Dumper:      s.logger(c),
481         }
482         s.stub.serveCurrentUserAdmin()
483         s.stub.serveFooBarFileCollections()
484         s.stub.serveKeepServices(stubServices)
485         s.stub.serveKeepstoreMounts()
486         s.stub.serveKeepstoreIndexFoo4Bar1()
487         trashReqs := s.stub.serveKeepstoreTrash()
488         pullReqs := s.stub.serveKeepstorePull()
489
490         stop := make(chan interface{})
491         s.config.RunPeriod = arvados.Duration(time.Millisecond)
492         srv, err := NewServer(s.config, opts)
493         c.Assert(err, check.IsNil)
494
495         done := make(chan bool)
496         go func() {
497                 srv.RunForever(stop)
498                 close(done)
499         }()
500
501         // Each run should send 4 pull lists + 4 trash lists. The
502         // first run should also send 4 empty trash lists at
503         // startup. We should complete all four runs in much less than
504         // a second.
505         for t0 := time.Now(); pullReqs.Count() < 16 && time.Since(t0) < 10*time.Second; {
506                 time.Sleep(time.Millisecond)
507         }
508         stop <- true
509         <-done
510         c.Check(pullReqs.Count() >= 16, check.Equals, true)
511         c.Check(trashReqs.Count(), check.Equals, pullReqs.Count()+4)
512         c.Check(s.getMetrics(c, srv), check.Matches, `(?ms).*\narvados_keepbalance_changeset_compute_seconds_count `+fmt.Sprintf("%d", pullReqs.Count()/4)+`\n.*`)
513 }
514
515 func (s *runSuite) getMetrics(c *check.C, srv *Server) string {
516         resp, err := http.Get("http://" + srv.listening + "/metrics")
517         c.Assert(err, check.IsNil)
518         c.Check(resp.StatusCode, check.Equals, http.StatusUnauthorized)
519
520         resp, err = http.Get("http://" + srv.listening + "/metrics?api_token=xyzzy")
521         c.Assert(err, check.IsNil)
522         c.Check(resp.StatusCode, check.Equals, http.StatusOK)
523         buf, err := ioutil.ReadAll(resp.Body)
524         c.Check(err, check.IsNil)
525         return string(buf)
526 }