[ https://issues.apache.org/jira/browse/BEAM-9746?focusedWorklogId=422386&page=com.atlassian.jira.plugin.system.issuetabpanels:worklog-tabpanel#worklog-422386 ]
ASF GitHub Bot logged work on BEAM-9746: ---------------------------------------- Author: ASF GitHub Bot Created on: 14/Apr/20 22:16 Start Date: 14/Apr/20 22:16 Worklog Time Spent: 10m Work Description: lostluck commented on pull request #11413: [BEAM-9746] check for 0 length copies from state URL: https://github.com/apache/beam/pull/11413#discussion_r408468603 ########## File path: sdks/go/pkg/beam/core/runtime/harness/statemgr_test.go ########## @@ -258,6 +261,167 @@ func TestStateChannel(t *testing.T) { } } +// TestStateKeyReader validates ordinary Read cases +func TestStateKeyReader(t *testing.T) { + const readLen = 4 + tests := []struct { + name string + buflens []int // sizes of the buffers received on the state channel. + numReads int + closed bool // tries to read from closed reader + noGet bool // tries to read from nil get response reader + }{ + { + name: "emptyData", + buflens: []int{-1}, + numReads: 1, + }, { + name: "singleBufferSingleRead", + buflens: []int{readLen}, + numReads: 2, + }, { + name: "singleBufferMultipleReads", + buflens: []int{2 * readLen}, + numReads: 3, + }, { + name: "singleBufferShortRead", + buflens: []int{readLen - 1}, + numReads: 2, + }, { + name: "multiBuffer", + buflens: []int{readLen, readLen}, + numReads: 3, + }, { + name: "multiBuffer-short-reads", + buflens: []int{readLen - 1, readLen - 1, readLen - 2}, + numReads: 4, + }, { + name: "emptyDataFirst", // Shouldn't happen, but not unreasonable to handle. + buflens: []int{-1, readLen, readLen}, + numReads: 4, + }, { + name: "emptyDataMid", // Shouldn't happen, but not unreasonable to handle. + buflens: []int{readLen, readLen, -1, readLen}, + numReads: 5, + }, { + name: "emptyDataLast", // Shouldn't happen, but not unreasonable to handle. + buflens: []int{readLen, readLen, -1}, + numReads: 3, + }, { + name: "emptyDataLast-short", + buflens: []int{3*readLen - 2, -1}, + numReads: 4, + }, { + name: "closed", + buflens: []int{-1, -1}, + numReads: 1, + closed: true, + }, { + name: "noGet", + buflens: []int{-1}, + numReads: 1, + noGet: true, + }, + } + for _, test := range tests { + t.Run(test.name, func(t *testing.T) { + ctx, cancelFn := context.WithCancel(context.Background()) + ch := &StateChannel{ + id: "test", + requests: make(chan *fnpb.StateRequest), + responses: make(map[string]chan<- *fnpb.StateResponse), + cancelFn: cancelFn, + DoneCh: ctx.Done(), + } + + // Handle the channel behavior asynchronously. + go func() { + for i := 0; i < len(test.buflens); i++ { + token := []byte(strconv.Itoa(i)) + var buf []byte + if test.buflens[i] >= 0 { + buf = bytes.Repeat([]byte{42}, test.buflens[i]) + } + // On the last request response pair, send no token. + if i+1 == len(test.buflens) { + token = nil + } + + req := <-ch.requests + + if test.noGet { + ch.responses[req.Id] <- &fnpb.StateResponse{ + Id: req.Id, + } + return + } + + ch.responses[req.Id] <- &fnpb.StateResponse{ + Id: req.Id, + Response: &fnpb.StateResponse_Get{ + Get: &fnpb.StateGetResponse{ + ContinuationToken: token, + Data: buf, + }, + }, + } + } + }() + + r := stateKeyReader{ + ch: ch, + } + + if test.closed { + err := r.Close() + if err != nil { + t.Errorf("unexpected error on Close(), got %v", err) + } + } + + var totalBytes int + for _, l := range test.buflens { + if l > 0 { + totalBytes += l + } + } + var finalerr error + var count, reads int + + // Read all the bytes. + for count <= totalBytes { + reads++ + b := make([]byte, readLen) // io.Read is keyed off of length, not capacity. + n, err := r.Read(b) + if err != nil { + finalerr = err + break + } + count += n + // Special check to avoid spurious zero elements. + if count == totalBytes && n == 0 { + t.Error("expected byte count read, last read is 0, but no EOF") + } + } + if got, want := reads, test.numReads; got != want { Review comment: No, it's my own personal habit after making mistakes on the ordering one to many times when debugging new tests. Since I've been working on beam for a bit, it's wherever I've been working on it. This way I never need to think of which one is 'got' or 'want', it's entirely unambiguous. It also simplifies copying the boiler plate for many sequential checks, where such leftover mistakes are common, as the variables only need to be updated in a single place, in the if definition scope. ---------------------------------------------------------------- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org Issue Time Tracking ------------------- Worklog Id: (was: 422386) Time Spent: 1h 40m (was: 1.5h) > [Go SDK] Empty side inputs causing spurious zero elements > --------------------------------------------------------- > > Key: BEAM-9746 > URL: https://issues.apache.org/jira/browse/BEAM-9746 > Project: Beam > Issue Type: Improvement > Components: sdk-go > Reporter: Robert Burke > Assignee: Robert Burke > Priority: Major > Time Spent: 1h 40m > Remaining Estimate: 0h > > A user discovered that empty side inputs would spuriously provide a single > zero element. > The error was narrowed down to the Go SDK's state manager codeĀ copying the > stateGetResponse data wasn't checking that the original data source even had > any bytes in it, leading it in particular to interpret length prefixed data > as having 0 length, which would cause zero value elements to be generated. > Notably, this caused empty strings. -- This message was sent by Atlassian Jira (v8.3.4#803005)