Add a script to copy eval run results to Turso (#2452)

This commit is contained in:
Chris Estreich 2025-04-09 21:18:07 -07:00 committed by GitHub
parent 5fa555e60d
commit 4e716263d0
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
3 changed files with 102 additions and 2 deletions

View file

@ -15,7 +15,8 @@
"db:check": "pnpm drizzle-kit check",
"db:up": "pnpm drizzle-kit up",
"db:studio": "pnpm drizzle-kit studio",
"db:enable-wal": "dotenvx run -f ../../.env -- tsx scripts/enable-wal.mts"
"db:enable-wal": "dotenvx run -f ../../.env -- tsx scripts/enable-wal.mts",
"db:copy-run": "dotenvx run -f ../../.env -- tsx scripts/copy-run.mts"
},
"dependencies": {
"@evals/types": "workspace:^",

View file

@ -0,0 +1,97 @@
import { drizzle } from "drizzle-orm/libsql"
import { eq } from "drizzle-orm"
import { db as sourceDb } from "../src/db.js"
import { schema } from "../src/schema.js"
const copyRun = async (runId: number) => {
const destDb = drizzle({
schema,
connection: { url: process.env.TURSO_CONNECTION_URL!, authToken: process.env.TURSO_AUTH_TOKEN! },
})
const run = await sourceDb.query.runs.findFirst({
where: eq(schema.runs.id, runId),
with: { taskMetrics: true },
})
if (!run) {
throw new Error(`Run with ID ${runId} not found in source database`)
}
if (!run.taskMetrics) {
throw new Error("Run is not completed")
}
console.log(`Copying run ${run.id}`)
// eslint-disable-next-line @typescript-eslint/no-unused-vars
const { id: _, ...runTaskMetricsValues } = run.taskMetrics
const [newRunTaskMetrics] = await destDb.insert(schema.taskMetrics).values(runTaskMetricsValues).returning()
if (!newRunTaskMetrics) {
throw new Error("Failed to insert run taskMetrics")
}
// eslint-disable-next-line @typescript-eslint/no-unused-vars
const { id: __, ...runValues } = run
const [newRun] = await destDb
.insert(schema.runs)
.values({ ...runValues, taskMetricsId: newRunTaskMetrics.id })
.returning()
if (!newRun) {
throw new Error("Failed to insert run")
}
const tasks = await sourceDb.query.tasks.findMany({
where: eq(schema.tasks.runId, run.id),
with: { taskMetrics: true },
})
console.log(`Copying ${tasks.length} tasks`)
for (const task of tasks) {
// eslint-disable-next-line @typescript-eslint/no-unused-vars
const { id: _, ...newTaskMetricsValues } = task.taskMetrics!
const [newTaskMetrics] = await destDb.insert(schema.taskMetrics).values(newTaskMetricsValues).returning()
if (!newTaskMetrics) {
throw new Error(`Failed to insert taskMetrics for task ${task.id}`)
}
// eslint-disable-next-line @typescript-eslint/no-unused-vars
const { id: __, ...newTaskValues } = task
const [newTask] = await destDb
.insert(schema.tasks)
.values({ ...newTaskValues, runId: newRun.id, taskMetricsId: newTaskMetrics.id })
.returning()
if (!newTask) {
throw new Error(`Failed to insert task ${task.id}`)
}
}
console.log(`Successfully copied run ${runId} with ${tasks.length} tasks`)
}
const main = async () => {
const runId = parseInt(process.argv[2], 10)
if (isNaN(runId)) {
console.error("Run ID must be a number")
process.exit(1)
}
try {
await copyRun(runId)
process.exit(0)
} catch (error) {
console.error(error)
process.exit(1)
}
}
main()

View file

@ -9,7 +9,9 @@
"OPENROUTER_MODEL_ID",
"PROMPT_PATH",
"WORKSPACE_PATH",
"BENCHMARKS_DB_PATH"
"BENCHMARKS_DB_PATH",
"TURSO_CONNECTION_URL",
"TURSO_AUTH_TOKEN"
],
"tasks": {
"lint": {},