Skip to content

Open-source AI coworkers that each get a computer of their own: a browser, files and tools, with every action decided before it happens and recorded after. Bring any AG-UI agent.

Record a computer reset as soon as the profile is gone (#404)

kevind97641e6171f
diff --git a/CHANGELOG.md b/CHANGELOG.md
index a28d4873..165d710f 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -8,6 +8,14 @@ Newest first. `Unreleased` is what is on `main` and not yet tagged.
## Unreleased
+### Wiping a computer is recorded even if clearing its stored state fails
+
+Resetting a computer destroys the profile first and wrote the audit row last, after two Postgres
+deletes. A connection reset, a failover or a statement timeout in either delete threw before the row
+was written, so the most destructive button in the product could leave a wiped computer -- every
+login gone, no undo -- with nothing on the trail to say who wiped it or when. The row is now written
+as soon as the profile is gone, which is the point after which nothing can be put back. A failure in
+either delete is still reported to the caller.
### Pressing Stop is recorded as a stop, not as a computer that is not running
The computer transport answered a Stop correctly only when it arrived before the request left. The
diff --git a/server/src/computer/gateway.ts b/server/src/computer/gateway.ts
index 21c7d985..99f0cf09 100644
--- a/server/src/computer/gateway.ts
+++ b/server/src/computer/gateway.ts
@@ -713,6 +713,22 @@ export function createComputerGateway(
*/
async resetComputer(botId: string, actor: ActionActor) {
const result = await provider.reset(botId);
+ /*
+ * The row goes in HERE, before the two deletes below, because this line is the point of no
+ * return: the profile is already gone and nothing after it can put the logins back.
+ *
+ * Both clears are Postgres deletes, and a connection reset, a failover or a statement timeout
+ * in either used to throw before the row was written -- leaving a computer wiped with nothing
+ * on the trail to say who wiped it, which is the one outcome the note above rules out. The
+ * failure still propagates, so the caller is told the clears did not finish.
+ */
+ await writeControlEvent(auditStore, "computer.reset", {
+ botId,
+ actor,
+ reason: result.cleared
+ ? "the computer and its saved state were deleted"
+ : "no saved state was present to delete",
+ });
// The refs the last snapshot handed out describe a page that no longer exists, and a fresh
// computer counts generations from one again, so the row has to go with the profile.
await snapshots.clear(botId);
@@ -725,13 +741,6 @@ export function createComputerGateway(
* anything a person would recognise as private.
*/
await pageFrames?.clear(botId);
- await writeControlEvent(auditStore, "computer.reset", {
- botId,
- actor,
- reason: result.cleared
- ? "the computer and its saved state were deleted"
- : "no saved state was present to delete",
- });
return result;
},
diff --git a/server/tests/computer-gateway.test.ts b/server/tests/computer-gateway.test.ts
index e2b4c255..75518a69 100644
--- a/server/tests/computer-gateway.test.ts
+++ b/server/tests/computer-gateway.test.ts
@@ -687,6 +687,67 @@ describe("the computer gateway", () => {
expect(rows[0]?.targetId).toBe("bot-2");
});
+ /*
+ * "The most destructive button we have. Every login the Bot had is gone and no undo exists, so the
+ * row is written whatever happens next."
+ *
+ * The profile is destroyed by `provider.reset` before either of the two Postgres deletes runs, so
+ * a delete that fails cannot put it back -- it can only take the row with it, which is the one
+ * thing that note rules out.
+ */
+ test("resetComputer records the reset even when clearing the stored page fails", async () => {
+ const { provider, fetchImpl } = fakeComputer({
+ resetResult: { cleared: true },
+ });
+ const { store, rows } = fakeAudit();
+ const snapshots: SnapshotStore = {
+ ...createInMemorySnapshotStore(),
+ clear: async () => {
+ throw new Error("connection reset by peer");
+ },
+ };
+ const gateway = createComputerGateway({
+ provider,
+ fetchImpl,
+ auditStore: store,
+ policy: () => PERMISSIVE,
+ snapshots,
+ });
+
+ await expect(gateway.resetComputer("bot-1", ACTOR)).rejects.toThrow(
+ "connection reset by peer",
+ );
+
+ expect(rows.map((row) => row.eventType)).toContain("computer.reset");
+ expect(rows[0]?.targetId).toBe("bot-1");
+ });
+
+ test("resetComputer records the reset even when clearing the screenshots fails", async () => {
+ const { provider, fetchImpl } = fakeComputer({
+ resetResult: { cleared: true },
+ });
+ const { store, rows } = fakeAudit();
+ const gateway = createComputerGateway({
+ provider,
+ fetchImpl,
+ auditStore: store,
+ policy: () => PERMISSIVE,
+ pageFrames: {
+ clear: async () => {
+ throw new Error("statement timeout");
+ },
+ } as unknown as NonNullable<
+ Parameters<typeof createComputerGateway>[0]["pageFrames"]
+ >,
+ });
+
+ await expect(gateway.resetComputer("bot-1", ACTOR)).rejects.toThrow(
+ "statement timeout",
+ );
+
+ expect(rows.map((row) => row.eventType)).toContain("computer.reset");
+ });
+
test("computers maps provider status 'running' and 'stopped' directly and preserves egress distinctions", async () => {
const locations: ComputerLocation[] = [
{