Start a local Postgres with wal2json using Docker for OLake testing. Configure replication slot, connect OLake UI, run syncs, and troubleshoot.

Learn how to connect containers to the host, across containers, or through proxies and VPNs in Docker Desktop.

<p><strong>Description</strong><br><br><code>types/state.go</code> holds the sync resume logic — backfill chunks, incremental cursors, the CDC global position, and <code>state.json</code> serialization — but had no test coverage, unlike <code>resolver</code>, <code>flattener</code> and <code>catalog</code>.<br>Bugs here fail silently: a wrong answer produces no error and no log line, it just re-copies data that was already synced or skips data that was not. This PR adds <strong>108 table-driven cases</strong> covering every function in the file.<br>Tests only — <strong>no production code is changed.</strong><br>Fixes #659<br>cc <a href="https://github.com/vaibhav-datazip">@vaibhav-datazip</a> <a href="https://github.com/nayanj98">@nayanj98</a><br><br><strong>Structure</strong><br><br>Follows the architecture of the existing unit tests, <code>reformat_test.go</code> in particular:<br>• one <code>Test&lt;Function&gt;</code> per function under test — <strong>all 18 functions in <code>state.go</code> map 1:1</strong><br>• a doc comment on every test function<br>• a flat, data-only <code>tests</code> table with <code>// ===== section =====</code> grouping and a single loop at the end<br>• <code>testify/assert</code> throughout, with <code>require</code> where continuing past a failure would nil-panic instead of reporting cleanly<br>• <code>t.Cleanup</code> restores the viper state path that the fixture overrides<br><br><strong>What is covered</strong><br><br>• <strong>Lifecycle</strong> — <code>SetType</code>, <code>isZero</code>, <code>ResetStreams</code><br>• <strong>Global state</strong> — <code>GetGlobal</code> / <code>SetGlobal</code>, including that <code>SetGlobal(nil, streamID)</code> (as called from <code>cdc.go:65</code>) preserves the stored position while still tracking the stream, and that a stream set is created when the decoded global has none<br>• <strong>Cursors</strong> — <code>GetCursor</code> / <code>SetCursor</code> / <code>ResetCursor</code>: empty-key no-ops, implicit <code>StreamState</code> creation on first write, primary vs. secondary cursor reset, and that streams are keyed on <strong>(namespace, name)</strong> together so the same table name in two schemas does not alias<br>• <strong>Chunks</strong> — <code>GetChunks</code> / <code>SetChunks</code> / <code>RemoveChunk</code>: the deliberate <code>FULLREFRESH</code> no-op, empty-set vs. no-chunks-recorded, and <code>RemoveChunk</code>'s <code>-1</code>-on-unknown-stream vs. remaining-count contract (the <code>0</code> return is what signals backfill completion at <code>backfill.go:61</code>)<br>• <strong><code>HasCompletedBackfill</code></strong> — exercised down both branches, since they are different code paths: <code>GLOBAL</code> (set membership) and <code>STREAM</code> (zero chunks remaining)<br>• <strong>Serialization</strong> — <code>State.MarshalJSON</code> dropping streams with <code>HoldsValue == false</code>; <code>StreamState.MarshalJSON</code> flattening the <code>sync.Map</code>; <code>StreamState.UnmarshalJSON</code> rehydrating a chunk array into a real <code>**Set[Chunk]</code>, asserted by type assertion rather than just by length<br>• <strong>Concurrency</strong> — 240 goroutines across the <code>RWMutex</code>, verified under <code>-race</code><br>• <strong>Current-behaviour pins</strong> — <code>initStreamState</code> leaves <code>StreamState.SyncMode</code> empty, so a freshly created stream state serializes with <code>"sync_mode": ""</code>. Pinned so that changing it is a deliberate act rather than an accident.<br><br><strong>Two notes for anyone writing tests against this file</strong><br><br>Both are documented in a comment at the top of the new file, since they cost real debugging time:<br>1. <code>State</code> embeds <code>**sync.RWMutex</code> — a <strong>pointer</strong> — so a bare <code>&amp;State{}</code> compiles fine and then nil-panics on every locking method. Fixtures need <code>&amp;State{RWMutex: &amp;sync.RWMutex{}}</code>.<br>2. Every locking mutator — <code>ResetStreams</code>, <code>ResetCursor</code>, <code>SetGlobal</code>, <code>SetCursor</code>, <code>SetChunks</code>, <code>RemoveChunk</code> (<code>SetType</code> is the exception: it neither locks nor logs) — ends in <code>LogState()</code>, which calls <code>logger.Fatalf</code> → <code>os.Exit(1)</code> when <code>constants.StatePath</code> is unset. That kills the test binary with <strong>no failure output at all</strong>. It is latent, too: <code>LogState</code> returns early while <code>isZero()</code> is true, so tests pass right up until one stores real data. Fixtures point the path at a <code>t.TempDir()</code> file.<br><br><strong>Type of change</strong><br><br>• Test coverage only (no production code changed, no behaviour change)<br>• This change requires a documentation update<br><br><strong>How Has This Been Tested?</strong><br><br>• <code>go test ./types/... -count=1 -v</code> — 108 cases, all passing<br>• <code>go test ./types/... -race</code> — clean, no data races<br>• <code>make test.unit</code> equivalent across the root module — all packages pass<br>• <code>make olake.lint</code> — exit 0, no findings<br>• <code>make build</code> — clean<br>• <code>gofmt -l -s</code> and <code>go vet ./types/...</code> — clean<br><strong>Coverage of <code>state.go</code> after this PR</strong> — every function is exercised; 15 of 18 are at 100%:<br><br>| Function | Coverage |<br>| ----------------------------------------------------------- | -------- |<br>| initStreamState, isZero, SetType, ResetStreams, ResetCursor | 100% |<br>| HasCompletedBackfill, GetGlobal, SetGlobal | 100% |<br>| GetCursor, SetCursor | 100% |<br>| GetChunks, SetChunks, RemoveChunk | 100% |<br>| State.MarshalJSON, LogWithLock | 100% |<br>| LogState | 83.3% |<br>| StreamState.MarshalJSON | 88.9% |<br>| StreamState.UnmarshalJSON | 84.2% |<br><br>The three partial ones are short only of branches that cannot be reached from a test:<br>• <code>LogState</code> — the <code>logger.Fatalf</code> → <code>os.Exit(1)</code> arm, which terminates the test binary by definition<br>• <code>StreamState.MarshalJSON</code> — the non-string <code>sync.Map</code> key arm, unreachable because every writer in <code>state.go</code> keys on a string<br>• <code>StreamState.UnmarshalJSON</code> — the two <code>json.Marshal</code>/<code>json.Unmarshal</code> error arms on a value that was just successfully decoded<br><br><strong>Screenshots or Recordings</strong><br><br>Uploading Rishyx_Shot 2026-08-18 at 12.01.07 PM.mov…<br><br><strong>Documentation</strong><br><br>• N/A (test changes only)<br><br><strong>Related PR's (If Any):</strong><br><br>None.<br><br>---<br><br><strong>One thing I would like a maintainer's call on</strong><br><br>While writing these tests I found that <code>RemoveChunk</code> does not check its type assertion:<br><br>// types/state.go:248<br>castedStateChunks, _ := stateChunks.(**Set[Chunk])<br>castedStateChunks.Remove(chunk) // nil pointer dereference if the assertion failed<br><br>This is reachable from a malformed <code>state.json</code>. <code>StreamState.UnmarshalJSON</code> only converts <code>chunks</code> into a <code>Set</code> when the JSON value is an <strong>array</strong> — a state file containing <code>"chunks": {"min": 1, "max": 10}</code> leaves a <code>map[string]interface{}</code> under that key instead. <code>GetChunks</code> handles that safely (it uses the comma-ok form and returns <code>nil</code>); <code>RemoveChunk</code> panics.<br><code>TestStreamStateUnmarshalJSON</code> documents the decode side of this — the case <em>"a chunk object is not converted into a chunk set"</em> asserts that a non-array value does not become a <code>**Set[Chunk]</code> — but it deliberately stops short of calling <code>RemoveChunk</code> on it, since that would assert a panic as intended behaviour.<br>I raised this on the issue before starting and did not get a call either way, so <strong>I have kept this PR tests-only</strong>. Happy to either add a small guard here or open a separate issue for it — just let me know which you prefer.</p>