| 1 | import { assert, describe, it } from "@effect/vitest" |
| 2 | import { Effect, Layer } from "effect" |
| 3 | import { type Backend, make } from "./harness.ts" |
| 4 | |
| 5 | describe("cluster lock-storage integration", () => { |
| 6 | for (const backend of ["pg", "mysql"] satisfies ReadonlyArray<Backend>) { |
| 7 | it.live( |
| 8 | `${backend}: rebuilds a blackholed reserved lock connection and rejoins without split brain`, |
| 9 | () => |
| 10 | Effect.gen(function*() { |
| 11 | const cluster = yield* make({ backend, entities: Layer.empty, config: { shardsPerGroup: 4 } }) |
| 12 | const [faulted] = yield* cluster.start(1) |
| 13 | yield* cluster.waitForStableAssignments() |
| 14 | const faultedAddress = `${faulted.address.host}:${faulted.address.port}` |
| 15 | |
| 16 | yield* cluster.faultLock(faulted, "blackhole") |
| 17 | const [peer] = yield* cluster.start(1) |
| 18 | const peerAddress = `${peer.address.host}:${peer.address.port}` |
| 19 | yield* cluster.waitUntil( |
| 20 | "The peer did not take over shards after the lock connection was blackholed", |
| 21 | Effect.gen(function*() { |
| 22 | yield* cluster.faultLock(faulted, "blackhole") |
| 23 | const assignments = Object.values(cluster.assignmentMap()) |
| 24 | assert(assignments.every((owners) => owners.length <= 1), "A shard had multiple owners") |
| 25 | return assignments.every((owners) => !owners.includes(faultedAddress)) && |
| 26 | assignments.some((owners) => owners.includes(peerAddress)) |
| 27 | }) |
| 28 | ) |
| 29 | |
| 30 | yield* cluster.faultLock(faulted, "clear") |
| 31 | const recovered = yield* cluster.waitForStableAssignments() |
| 32 | assert( |
| 33 | Object.values(recovered).some((owners) => owners.includes(faultedAddress)), |
| 34 | "The recovered runner did not rejoin the assignment ring" |
| 35 | ) |
| 36 | }) |
| 37 | ) |
| 38 | |
| 39 | it.live(`${backend}: does not reacquire a shard before forced release completes`, () => |
| 40 | Effect.gen(function*() { |
| 41 | const cluster = yield* make({ backend, entities: Layer.empty, config: { shardsPerGroup: 4 } }) |
| 42 | const [faulted] = yield* cluster.start(1) |
| 43 | yield* cluster.waitForStableAssignments() |
| 44 | const faultedAddress = `${faulted.address.host}:${faulted.address.port}` |
| 45 | |
| 46 | yield* cluster.faultLock(faulted, "hangRelease") |
| 47 | yield* cluster.waitUntil( |
| 48 | "The faulted runner did not drop its shards", |
| 49 | Effect.sync(() => { |
| 50 | const assignments = Object.values(cluster.assignmentMap()) |
| 51 | assert(assignments.every((owners) => owners.length <= 1), "A shard had multiple owners") |
| 52 | return assignments.every((owners) => !owners.includes(faultedAddress)) |
| 53 | }) |
| 54 | ) |
| 55 | |
| 56 | const [peer] = yield* cluster.start(1) |
| 57 | const peerAddress = `${peer.address.host}:${peer.address.port}` |
| 58 | yield* cluster.waitUntil( |
| 59 | "The peer did not make progress while the forced release was pending", |
| 60 | Effect.sync(() => { |
| 61 | const assignments = Object.values(cluster.assignmentMap()) |
| 62 | assert(assignments.every((owners) => owners.length <= 1), "A shard had multiple owners") |
| 63 | return assignments.some((owners) => owners.includes(peerAddress)) |
| 64 | }) |
| 65 | ) |
| 66 | |
| 67 | yield* cluster.faultLock(faulted, "clear") |
| 68 | let previous = "" |
| 69 | let stablePolls = 0 |
| 70 | yield* cluster.waitUntil( |
| 71 | "The cluster did not reach a terminal assignment after the forced release", |
| 72 | Effect.sync(() => { |
| 73 | const assignments = cluster.assignmentMap() |
| 74 | const owners = Object.values(assignments) |
| 75 | assert(owners.every((shardOwners) => shardOwners.length <= 1), "A shard had multiple owners") |
| 76 | const complete = owners.every((shardOwners) => shardOwners.length === 1) && |
| 77 | owners.some((shardOwners) => shardOwners.includes(faultedAddress)) && |
| 78 | owners.some((shardOwners) => shardOwners.includes(peerAddress)) |
| 79 | const encoded = complete ? JSON.stringify(assignments) : "" |
| 80 | stablePolls = encoded !== "" && encoded === previous ? stablePolls + 1 : 0 |
| 81 | previous = encoded |
| 82 | return stablePolls >= 3 |
| 83 | }) |
| 84 | ) |
| 85 | })) |
| 86 | |
| 87 | it.live(`${backend}: keeps peers progressing when a lock query is permanently stuck`, () => |
| 88 | Effect.gen(function*() { |
| 89 | const cluster = yield* make({ backend, entities: Layer.empty, config: { shardsPerGroup: 4 } }) |
| 90 | const [faulted] = yield* cluster.start(1) |
| 91 | yield* cluster.waitForStableAssignments() |
| 92 | const faultedAddress = `${faulted.address.host}:${faulted.address.port}` |
| 93 | |
| 94 | yield* cluster.faultLock(faulted, "stuck") |
| 95 | yield* cluster.waitUntil( |
| 96 | "The permanently stuck runner did not drop its shards", |
| 97 | Effect.sync(() => Object.values(cluster.assignmentMap()).every((owners) => !owners.includes(faultedAddress))) |
| 98 | ) |
| 99 | |
| 100 | const [peer] = yield* cluster.start(1) |
| 101 | const peerAddress = `${peer.address.host}:${peer.address.port}` |
| 102 | yield* cluster.waitUntil( |
| 103 | "The peer did not acquire shards while the other lock query remained stuck", |
| 104 | Effect.sync(() => { |
| 105 | const owners = Object.values(cluster.assignmentMap()) |
| 106 | assert(owners.every((shardOwners) => !shardOwners.includes(faultedAddress))) |
| 107 | return owners.some((shardOwners) => shardOwners.includes(peerAddress)) |
| 108 | }) |
| 109 | ) |
| 110 | |
| 111 | yield* cluster.faultLock(faulted, "clear") |
| 112 | yield* cluster.waitForStableAssignments() |
| 113 | })) |
| 114 | |
| 115 | it.live(`${backend}: converges after repeated lock-connection rebuilds and a hung release`, () => |
| 116 | Effect.gen(function*() { |
| 117 | const cluster = yield* make({ backend, entities: Layer.empty, config: { shardsPerGroup: 4 } }) |
| 118 | const [faulted] = yield* cluster.start(1) |
| 119 | yield* cluster.waitForStableAssignments() |
| 120 | const faultedAddress = `${faulted.address.host}:${faulted.address.port}` |
| 121 | |
| 122 | yield* cluster.faultLock(faulted, "fail") |
| 123 | let healthyPolls = 0 |
| 124 | yield* cluster.waitUntil( |
| 125 | "The runner lost shards while rebuilding failed lock connections", |
| 126 | Effect.gen(function*() { |
| 127 | yield* cluster.faultLock(faulted, "fail") |
| 128 | const assignments = Object.values(cluster.assignmentMap()) |
| 129 | assert( |
| 130 | assignments.every((owners) => owners.length === 1 && owners.includes(faultedAddress)), |
| 131 | "The runner lost a shard during lock-connection rebuild" |
| 132 | ) |
| 133 | healthyPolls++ |
| 134 | return healthyPolls >= 10 |
| 135 | }) |
| 136 | ) |
| 137 | |
| 138 | yield* cluster.faultLock(faulted, "hangRelease") |
| 139 | yield* cluster.waitUntil( |
| 140 | "The runner did not drop its shards while the previous connection release hung", |
| 141 | Effect.sync(() => Object.values(cluster.assignmentMap()).every((owners) => !owners.includes(faultedAddress))) |
| 142 | ) |
| 143 | |
| 144 | const [peer] = yield* cluster.start(1) |
| 145 | const peerAddress = `${peer.address.host}:${peer.address.port}` |
| 146 | yield* cluster.waitUntil( |
| 147 | "The peer did not acquire shards during the repeated rebuild", |
| 148 | Effect.sync(() => Object.values(cluster.assignmentMap()).some((owners) => owners.includes(peerAddress))) |
| 149 | ) |
| 150 | |
| 151 | yield* cluster.faultLock(faulted, "clear") |
| 152 | yield* cluster.waitForStableAssignments() |
| 153 | })) |
| 154 | } |
| 155 | }) |