fix: resolve alignment errors and CI timeouts in conformance tests

Use DataView instead of typed arrays in encodeColumnarRun to avoid
Float64Array/BigInt64Array alignment requirements. Fix HashJoin batch
tail loss when memory budget exceeded mid-batch. Increase vitest
hook/test timeout to 300s for 5M-row conformance tests on CI runners.

Author: teamchong

src/operators.ts

@@ -1174,17 +1174,21 @@ export class HashJoinOperator implements Operator {
       const inMemoryRows: Row[] = [...firstBatch];
       this.buildSizeBytes = batchSizeBytes;
       let exceeds = false;
+      let exceededBatchTail: Row[] = [];
 
       while (true) {
         const batch = await this.right.next();
         if (!batch) break;
-        for (const row of batch) {
+        for (let ri = 0; ri < batch.length; ri++) {
+          const row = batch[ri];
           const rowSize = estimateRowSize(row);
           this.buildSizeBytes += rowSize;
           inMemoryRows.push(row);
 
           if (this.spill && this.buildSizeBytes > this.memoryBudget) {
             exceeds = true;
+            // Capture remaining rows in this batch that we haven't visited
+            exceededBatchTail = batch.slice(ri + 1);
             break;
           }
         }
@@ -1209,7 +1213,11 @@ export class HashJoinOperator implements Operator {
       await this.partitionRightRows(inMemoryRows);
       inMemoryRows.length = 0;
 
-      // Continue consuming remaining right-side rows
+      // Partition the remaining rows from the batch that triggered the exceed
+      if (exceededBatchTail.length > 0) {
+        await this.partitionRightRows(exceededBatchTail);
+      }
+      // Continue consuming remaining right-side batches
       await this.consumeRemainingRight();
       return;
     }

src/r2-spill.ts

@@ -219,19 +219,17 @@ export function encodeColumnarRun(rows: Row[]): ArrayBuffer {
 
     switch (dtype) {
       case DTYPE_F64: {
-        const arr = new Float64Array(buf, offset, rowCount);
         for (let ri = 0; ri < rowCount; ri++) {
           const val = rows[ri][name];
-          arr[ri] = typeof val === "number" ? val : 0;
+          view.setFloat64(offset + ri * 8, typeof val === "number" ? val : 0, true);
         }
         offset += rowCount * 8;
         break;
       }
       case DTYPE_I64: {
-        const arr = new BigInt64Array(buf, offset, rowCount);
         for (let ri = 0; ri < rowCount; ri++) {
           const val = rows[ri][name];
-          arr[ri] = typeof val === "bigint" ? val : 0n;
+          view.setBigInt64(offset + ri * 8, typeof val === "bigint" ? val : 0n, true);
         }
         offset += rowCount * 8;
         break;
@@ -257,14 +255,13 @@ export function encodeColumnarRun(rows: Row[]): ArrayBuffer {
 
         view.setUint32(offset, totalLen, true); offset += 4;
 
-        // Offsets array
-        const offsets = new Uint32Array(buf, offset, rowCount + 1);
+        // Offsets array (use DataView to avoid alignment issues)
         let strOff = 0;
         for (let ri = 0; ri < rowCount; ri++) {
-          offsets[ri] = strOff;
+          view.setUint32(offset + ri * 4, strOff, true);
           strOff += encodedStrs[ri].length;
         }
-        offsets[rowCount] = strOff;
+        view.setUint32(offset + rowCount * 4, strOff, true);
         offset += (rowCount + 1) * 4;
 
         // String data

vitest.config.ts

@@ -10,5 +10,7 @@ export default defineConfig({
   test: {
     environment: "node",
     setupFiles: ["./src/test-setup.ts"],
+    hookTimeout: 300_000,
+    testTimeout: 300_000,
   },
 });