fix(eval): coalesced python kernel replacement
Tracked retained Python kernel generations and shared one replacement promise per dead generation. Reset and disposal now invalidate and drain replacement work before allowing a new session to take ownership. Added deterministic fake-kernel coverage for concurrent callers, cancellation, reset, owner/global disposal, and independent cwd keys. Fixes #6367
This commit is contained in:
@@ -4,6 +4,7 @@
|
||||
|
||||
### Fixed
|
||||
|
||||
- Fixed concurrent dead-kernel replacement in persistent Python sessions starting multiple generations and orphaning the losing process; callers now share one generation-scoped replacement, while resets and disposal invalidate and drain stale replacements ([#6367](https://github.com/can1357/oh-my-pi/issues/6367)).
|
||||
- Fixed credential-shaped tokens (GitHub/GitLab/OpenAI/Anthropic key patterns) being redacted from outbound provider requests even with `secrets.enabled` off; the pattern redaction now follows the `secrets.enabled` ("Hide Secrets") setting like the secret obfuscator.
|
||||
- Fixed Ctrl-clicking a wrapped OAuth authorization URL opening only the clicked row's truncated fragment by preserving the complete hyperlink target on every rendered row.
|
||||
- Fixed used-only absolute usage amounts across output surfaces: CLI now renders `$123.45 used`; the TUI shows a neutral, width-bounded amount instead of a pending/dotted/account-count placeholder; and ACP preserves `123.45 usd used` while suppressing duplicate window suffixes such as `— extra`. ([#5575](https://github.com/can1357/oh-my-pi/issues/5575))
|
||||
|
||||
@@ -21,6 +21,7 @@ import {
|
||||
type KernelDisplayOutput,
|
||||
type KernelExecuteOptions,
|
||||
type KernelExecuteResult,
|
||||
type KernelShutdownResult,
|
||||
PythonKernel,
|
||||
} from "./kernel";
|
||||
import { resolveExplicitPythonRuntime } from "./runtime";
|
||||
@@ -136,11 +137,19 @@ export interface PythonResult {
|
||||
// the same tuple; the kernel stays alive until the last owner detaches.
|
||||
// ---------------------------------------------------------------------------
|
||||
|
||||
interface SessionKernelReplacement {
|
||||
generation: number;
|
||||
deadlineMs?: number;
|
||||
promise: Promise<PythonKernel>;
|
||||
}
|
||||
|
||||
interface PythonSession {
|
||||
sessionKey: string;
|
||||
sessionId: string;
|
||||
cwd: string;
|
||||
kernel: PythonKernel;
|
||||
generation: number;
|
||||
replacement?: SessionKernelReplacement;
|
||||
ownerIds: Set<string>;
|
||||
hasFallbackOwner: boolean;
|
||||
}
|
||||
@@ -254,6 +263,7 @@ async function acquireSession(
|
||||
sessionId,
|
||||
cwd,
|
||||
kernel,
|
||||
generation: 0,
|
||||
ownerIds: new Set(),
|
||||
hasFallbackOwner: false,
|
||||
};
|
||||
@@ -272,31 +282,99 @@ async function acquireSession(
|
||||
|
||||
async function replaceSessionKernel(
|
||||
session: PythonSession,
|
||||
kernel: PythonKernel,
|
||||
generation: number,
|
||||
cwd: string,
|
||||
options: PythonExecutorOptions,
|
||||
): Promise<void> {
|
||||
const old = session.kernel;
|
||||
const remaining = getRemainingTimeoutMs(options.deadlineMs);
|
||||
await old
|
||||
.shutdown(remaining !== undefined ? { timeoutMs: Math.max(0, remaining) } : undefined)
|
||||
.catch(() => undefined);
|
||||
if (sessions.get(session.sessionKey) !== session) {
|
||||
): Promise<PythonKernel> {
|
||||
const inFlight = session.replacement;
|
||||
if (inFlight?.generation === generation) {
|
||||
if (
|
||||
inFlight.deadlineMs !== undefined &&
|
||||
(options.deadlineMs === undefined || options.deadlineMs > inFlight.deadlineMs)
|
||||
) {
|
||||
inFlight.deadlineMs = options.deadlineMs;
|
||||
}
|
||||
return await waitForPromiseWithCancellation(inFlight.promise, options, PythonExecutionCancelledError);
|
||||
}
|
||||
if (sessions.get(session.sessionKey) !== session || session.generation !== generation || session.kernel !== kernel) {
|
||||
throw new PythonExecutionCancelledError(false);
|
||||
}
|
||||
requireRemainingTimeoutMs(options.deadlineMs);
|
||||
const next = await startKernel(cwd, options);
|
||||
if (sessions.get(session.sessionKey) !== session) {
|
||||
await next.shutdown().catch(() => undefined);
|
||||
throw new PythonExecutionCancelledError(false);
|
||||
|
||||
const deferred = Promise.withResolvers<PythonKernel>();
|
||||
const replacement: SessionKernelReplacement = {
|
||||
generation,
|
||||
deadlineMs: options.deadlineMs,
|
||||
promise: deferred.promise,
|
||||
};
|
||||
session.replacement = replacement;
|
||||
void (async () => {
|
||||
try {
|
||||
const remaining = getRemainingTimeoutMs(options.deadlineMs);
|
||||
await kernel
|
||||
.shutdown(remaining !== undefined ? { timeoutMs: Math.max(0, remaining) } : undefined)
|
||||
.catch(() => undefined);
|
||||
if (replacement.deadlineMs !== undefined && replacement.deadlineMs <= Date.now()) {
|
||||
throw new PythonExecutionCancelledError(true);
|
||||
}
|
||||
if (
|
||||
sessions.get(session.sessionKey) !== session ||
|
||||
session.generation !== generation ||
|
||||
session.kernel !== kernel
|
||||
) {
|
||||
throw new PythonExecutionCancelledError(false);
|
||||
}
|
||||
const next = await startKernel(cwd, {
|
||||
...options,
|
||||
signal: undefined,
|
||||
deadlineMs: undefined,
|
||||
});
|
||||
if (
|
||||
sessions.get(session.sessionKey) !== session ||
|
||||
session.generation !== generation ||
|
||||
session.kernel !== kernel
|
||||
) {
|
||||
await next.shutdown().catch(() => undefined);
|
||||
throw new PythonExecutionCancelledError(false);
|
||||
}
|
||||
session.kernel = next;
|
||||
session.generation += 1;
|
||||
deferred.resolve(next);
|
||||
} catch (err) {
|
||||
deferred.reject(err);
|
||||
} finally {
|
||||
if (session.replacement === replacement) session.replacement = undefined;
|
||||
}
|
||||
})();
|
||||
return await waitForPromiseWithCancellation(deferred.promise, options, PythonExecutionCancelledError);
|
||||
}
|
||||
|
||||
async function shutdownInvalidatedSession(session: PythonSession): Promise<KernelShutdownResult> {
|
||||
const replacement = session.replacement;
|
||||
if (replacement) await replacement.promise.catch(() => undefined);
|
||||
return await session.kernel.shutdown();
|
||||
}
|
||||
|
||||
async function acquireLiveSessionKernel(
|
||||
session: PythonSession,
|
||||
cwd: string,
|
||||
options: PythonExecutorOptions,
|
||||
): Promise<PythonKernel> {
|
||||
while (sessions.get(session.sessionKey) === session) {
|
||||
const kernel = session.kernel;
|
||||
const generation = session.generation;
|
||||
if (kernel.isAlive()) return kernel;
|
||||
await replaceSessionKernel(session, kernel, generation, cwd, options);
|
||||
}
|
||||
session.kernel = next;
|
||||
throw new PythonExecutionCancelledError(false);
|
||||
}
|
||||
|
||||
async function resetSession(sessionKey: string): Promise<void> {
|
||||
const existing = sessions.get(sessionKey) ?? (await startingSessions.get(sessionKey)?.catch(() => undefined));
|
||||
if (!existing) return;
|
||||
existing.generation += 1;
|
||||
sessions.delete(sessionKey);
|
||||
await existing.kernel.shutdown().catch(() => undefined);
|
||||
await shutdownInvalidatedSession(existing).catch(() => undefined);
|
||||
}
|
||||
|
||||
// ---------------------------------------------------------------------------
|
||||
@@ -315,9 +393,10 @@ export async function disposeAllKernelSessions(): Promise<void> {
|
||||
}
|
||||
}
|
||||
for (const [id, session] of all) {
|
||||
session.generation += 1;
|
||||
if (sessions.get(id) === session) sessions.delete(id);
|
||||
}
|
||||
const results = await Promise.allSettled(all.map(([, session]) => session.kernel.shutdown()));
|
||||
const results = await Promise.allSettled(all.map(([, session]) => shutdownInvalidatedSession(session)));
|
||||
for (let i = 0; i < all.length; i += 1) {
|
||||
const [id, session] = all[i];
|
||||
const result = results[i];
|
||||
@@ -344,9 +423,10 @@ export async function disposeKernelSessionsByOwner(ownerId: string): Promise<voi
|
||||
session.ownerIds.delete(ownerId);
|
||||
}
|
||||
for (const session of toShutdown) {
|
||||
session.generation += 1;
|
||||
if (sessions.get(session.sessionKey) === session) sessions.delete(session.sessionKey);
|
||||
}
|
||||
const results = await Promise.allSettled(toShutdown.map(session => session.kernel.shutdown()));
|
||||
const results = await Promise.allSettled(toShutdown.map(session => shutdownInvalidatedSession(session)));
|
||||
for (let i = 0; i < toShutdown.length; i += 1) {
|
||||
const session = toShutdown[i];
|
||||
const result = results[i];
|
||||
@@ -458,31 +538,21 @@ async function executeOnSession(code: string, cwd: string, options: PythonExecut
|
||||
isTimedOutCancellation(options.signal.reason, PythonExecutionCancelledError, options.signal),
|
||||
);
|
||||
}
|
||||
if (sessions.get(session.sessionKey) !== session) {
|
||||
const kernel = await acquireLiveSessionKernel(session, cwd, options);
|
||||
if (sessions.get(session.sessionKey) !== session || session.kernel !== kernel) {
|
||||
throw new PythonExecutionCancelledError(false);
|
||||
}
|
||||
if (!session.kernel.isAlive()) {
|
||||
await replaceSessionKernel(session, cwd, options);
|
||||
if (sessions.get(session.sessionKey) !== session) {
|
||||
throw new PythonExecutionCancelledError(false);
|
||||
}
|
||||
}
|
||||
const runOptions = { ...options, cwd };
|
||||
try {
|
||||
return await executeWithKernel(session.kernel, code, runOptions);
|
||||
return await executeWithKernel(kernel, code, runOptions);
|
||||
} catch (err) {
|
||||
if (isCancellationError(err, PythonExecutionCancelledError) || options.signal?.aborted) throw err;
|
||||
if (session.kernel.isAlive()) throw err;
|
||||
if (sessions.get(session.sessionKey) !== session) {
|
||||
if (kernel.isAlive()) throw err;
|
||||
const retryKernel = await acquireLiveSessionKernel(session, cwd, options);
|
||||
if (sessions.get(session.sessionKey) !== session || session.kernel !== retryKernel) {
|
||||
throw new PythonExecutionCancelledError(false);
|
||||
}
|
||||
// Shared kernels are keyed by cwd, so a dead kernel can be recreated in place
|
||||
// without risking cross-directory state bleed.
|
||||
await replaceSessionKernel(session, cwd, options);
|
||||
if (sessions.get(session.sessionKey) !== session) {
|
||||
throw new PythonExecutionCancelledError(false);
|
||||
}
|
||||
return await executeWithKernel(session.kernel, code, runOptions);
|
||||
return await executeWithKernel(retryKernel, code, runOptions);
|
||||
}
|
||||
}
|
||||
|
||||
|
||||
@@ -1,5 +1,9 @@
|
||||
import { afterEach, describe, expect, it } from "bun:test";
|
||||
import { disposeAllKernelSessions, executePython } from "@oh-my-pi/pi-coding-agent/eval/py/executor";
|
||||
import {
|
||||
disposeAllKernelSessions,
|
||||
disposeKernelSessionsByOwner,
|
||||
executePython,
|
||||
} from "@oh-my-pi/pi-coding-agent/eval/py/executor";
|
||||
import {
|
||||
type KernelExecuteOptions,
|
||||
type KernelExecuteResult,
|
||||
@@ -29,6 +33,10 @@ class FakeKernel {
|
||||
return this.#alive;
|
||||
}
|
||||
|
||||
markDead(): void {
|
||||
this.#alive = false;
|
||||
}
|
||||
|
||||
async execute(code: string, options?: KernelExecuteOptions): Promise<KernelExecuteResult> {
|
||||
this.executeCalls.push(code);
|
||||
this.#onExecute?.(options);
|
||||
@@ -53,6 +61,10 @@ const okResult: KernelExecuteResult = {
|
||||
stdinRequested: false,
|
||||
};
|
||||
|
||||
async function flushMicrotasks(): Promise<void> {
|
||||
for (let i = 0; i < 5; i += 1) await Promise.resolve();
|
||||
}
|
||||
|
||||
describe("executePython session lifecycle", () => {
|
||||
const originalStart = PythonKernel.start;
|
||||
|
||||
@@ -95,6 +107,215 @@ describe("executePython session lifecycle", () => {
|
||||
expect(liveKernel.executeCalls).toEqual(["print('restart')"]);
|
||||
});
|
||||
|
||||
it("coalesces concurrent replacement of one dead session generation", async () => {
|
||||
const deadKernel = new FakeKernel(okResult);
|
||||
const replacementOne = new FakeKernel(okResult);
|
||||
const replacementTwo = new FakeKernel(okResult);
|
||||
const replacementStarted = Promise.withResolvers<void>();
|
||||
const releaseReplacement = Promise.withResolvers<void>();
|
||||
const replacements = [replacementOne, replacementTwo];
|
||||
let startCount = 0;
|
||||
|
||||
PythonKernel.start = async () => {
|
||||
startCount += 1;
|
||||
if (startCount === 1) return deadKernel as unknown as PythonKernel;
|
||||
replacementStarted.resolve();
|
||||
await releaseReplacement.promise;
|
||||
return replacements.shift() as unknown as PythonKernel;
|
||||
};
|
||||
|
||||
await executePython("print('setup')", { sessionId: "session-concurrent-restart" });
|
||||
deadKernel.markDead();
|
||||
|
||||
const first = executePython("print('first')", { sessionId: "session-concurrent-restart" });
|
||||
await replacementStarted.promise;
|
||||
const second = executePython("print('second')", { sessionId: "session-concurrent-restart" });
|
||||
await Promise.resolve();
|
||||
await Promise.resolve();
|
||||
|
||||
expect(startCount).toBe(2);
|
||||
|
||||
releaseReplacement.resolve();
|
||||
await Promise.all([first, second]);
|
||||
|
||||
expect(replacementOne.executeCalls).toEqual(["print('first')", "print('second')"]);
|
||||
expect(replacementTwo.executeCalls).toEqual([]);
|
||||
|
||||
await disposeAllKernelSessions();
|
||||
expect(replacementOne.shutdownCalls).toBe(1);
|
||||
expect(replacementTwo.shutdownCalls).toBe(0);
|
||||
});
|
||||
|
||||
it("keeps a shared replacement alive when one caller cancels", async () => {
|
||||
const deadKernel = new FakeKernel(okResult);
|
||||
const replacement = new FakeKernel(okResult);
|
||||
const replacementStarted = Promise.withResolvers<void>();
|
||||
const releaseReplacement = Promise.withResolvers<void>();
|
||||
let startCount = 0;
|
||||
|
||||
PythonKernel.start = async () => {
|
||||
startCount += 1;
|
||||
if (startCount === 1) return deadKernel as unknown as PythonKernel;
|
||||
replacementStarted.resolve();
|
||||
await releaseReplacement.promise;
|
||||
return replacement as unknown as PythonKernel;
|
||||
};
|
||||
|
||||
await executePython("print('setup')", { sessionId: "session-cancelled-restart" });
|
||||
deadKernel.markDead();
|
||||
|
||||
const abortController = new AbortController();
|
||||
const cancelled = executePython("print('cancelled')", {
|
||||
sessionId: "session-cancelled-restart",
|
||||
signal: abortController.signal,
|
||||
});
|
||||
await replacementStarted.promise;
|
||||
const retained = executePython("print('retained')", { sessionId: "session-cancelled-restart" });
|
||||
await flushMicrotasks();
|
||||
abortController.abort(Object.assign(new Error("replacement wait cancelled"), { name: "AbortError" }));
|
||||
|
||||
expect((await cancelled).cancelled).toBe(true);
|
||||
expect(startCount).toBe(2);
|
||||
|
||||
releaseReplacement.resolve();
|
||||
expect((await retained).cancelled).toBe(false);
|
||||
expect(replacement.executeCalls).toEqual(["print('retained')"]);
|
||||
});
|
||||
|
||||
it("invalidates an in-flight replacement before resetting to a fresh generation", async () => {
|
||||
const deadKernel = new FakeKernel(okResult);
|
||||
const staleReplacement = new FakeKernel(okResult);
|
||||
const freshKernel = new FakeKernel(okResult);
|
||||
const replacementStarted = Promise.withResolvers<void>();
|
||||
const releaseReplacement = Promise.withResolvers<void>();
|
||||
let startCount = 0;
|
||||
|
||||
PythonKernel.start = async () => {
|
||||
startCount += 1;
|
||||
if (startCount === 1) return deadKernel as unknown as PythonKernel;
|
||||
if (startCount === 2) {
|
||||
replacementStarted.resolve();
|
||||
await releaseReplacement.promise;
|
||||
return staleReplacement as unknown as PythonKernel;
|
||||
}
|
||||
return freshKernel as unknown as PythonKernel;
|
||||
};
|
||||
|
||||
await executePython("print('setup')", { sessionId: "session-reset-replacement" });
|
||||
deadKernel.markDead();
|
||||
|
||||
const obsolete = executePython("print('obsolete')", { sessionId: "session-reset-replacement" });
|
||||
await replacementStarted.promise;
|
||||
const reset = executePython("print('reset')", {
|
||||
sessionId: "session-reset-replacement",
|
||||
reset: true,
|
||||
});
|
||||
await flushMicrotasks();
|
||||
releaseReplacement.resolve();
|
||||
|
||||
expect((await obsolete).cancelled).toBe(true);
|
||||
expect((await reset).cancelled).toBe(false);
|
||||
expect(staleReplacement.executeCalls).toEqual([]);
|
||||
expect(staleReplacement.shutdownCalls).toBe(1);
|
||||
expect(freshKernel.executeCalls).toEqual(["print('reset')"]);
|
||||
|
||||
await executePython("print('later')", { sessionId: "session-reset-replacement" });
|
||||
expect(startCount).toBe(3);
|
||||
expect(freshKernel.executeCalls).toEqual(["print('reset')", "print('later')"]);
|
||||
});
|
||||
|
||||
it("drains replacements invalidated by owner and global disposal", async () => {
|
||||
const ownerKernel = new FakeKernel(okResult);
|
||||
const globalKernel = new FakeKernel(okResult);
|
||||
const ownerReplacement = new FakeKernel(okResult);
|
||||
const globalReplacement = new FakeKernel(okResult);
|
||||
const replacementsStarted = Promise.withResolvers<void>();
|
||||
const releaseReplacements = Promise.withResolvers<void>();
|
||||
const initialKernels = [ownerKernel, globalKernel];
|
||||
const replacementKernels = [ownerReplacement, globalReplacement];
|
||||
let replacementStartCount = 0;
|
||||
|
||||
PythonKernel.start = async () => {
|
||||
const initial = initialKernels.shift();
|
||||
if (initial) return initial as unknown as PythonKernel;
|
||||
replacementStartCount += 1;
|
||||
if (replacementStartCount === 2) replacementsStarted.resolve();
|
||||
await releaseReplacements.promise;
|
||||
return replacementKernels.shift() as unknown as PythonKernel;
|
||||
};
|
||||
|
||||
await executePython("print('owner setup')", {
|
||||
sessionId: "session-owner-disposal-replacement",
|
||||
kernelOwnerId: "replacement-owner",
|
||||
});
|
||||
await executePython("print('global setup')", { sessionId: "session-global-disposal-replacement" });
|
||||
ownerKernel.markDead();
|
||||
globalKernel.markDead();
|
||||
|
||||
const ownerExecution = executePython("print('owner obsolete')", {
|
||||
sessionId: "session-owner-disposal-replacement",
|
||||
kernelOwnerId: "replacement-owner",
|
||||
});
|
||||
const globalExecution = executePython("print('global obsolete')", {
|
||||
sessionId: "session-global-disposal-replacement",
|
||||
});
|
||||
await replacementsStarted.promise;
|
||||
|
||||
const ownerDisposal = disposeKernelSessionsByOwner("replacement-owner");
|
||||
await flushMicrotasks();
|
||||
const globalDisposal = disposeAllKernelSessions();
|
||||
await flushMicrotasks();
|
||||
releaseReplacements.resolve();
|
||||
|
||||
expect((await ownerExecution).cancelled).toBe(true);
|
||||
expect((await globalExecution).cancelled).toBe(true);
|
||||
await Promise.all([ownerDisposal, globalDisposal]);
|
||||
expect(ownerReplacement.executeCalls).toEqual([]);
|
||||
expect(globalReplacement.executeCalls).toEqual([]);
|
||||
expect(ownerReplacement.shutdownCalls).toBe(1);
|
||||
expect(globalReplacement.shutdownCalls).toBe(1);
|
||||
});
|
||||
|
||||
it("keeps replacement coordination independent across normalized cwd keys", async () => {
|
||||
const deadOne = new FakeKernel(okResult);
|
||||
const deadTwo = new FakeKernel(okResult);
|
||||
const replacementOne = new FakeKernel(okResult);
|
||||
const replacementTwo = new FakeKernel(okResult);
|
||||
const kernels = [deadOne, deadTwo, replacementOne, replacementTwo];
|
||||
let startCount = 0;
|
||||
|
||||
PythonKernel.start = async () => {
|
||||
startCount += 1;
|
||||
return kernels.shift() as unknown as PythonKernel;
|
||||
};
|
||||
|
||||
await executePython("print('setup one')", {
|
||||
cwd: "/tmp/replacement-key-one",
|
||||
sessionId: "session-independent-replacement",
|
||||
});
|
||||
await executePython("print('setup two')", {
|
||||
cwd: "/tmp/replacement-key-two",
|
||||
sessionId: "session-independent-replacement",
|
||||
});
|
||||
deadOne.markDead();
|
||||
deadTwo.markDead();
|
||||
|
||||
await Promise.all([
|
||||
executePython("print('one')", {
|
||||
cwd: "/tmp/replacement-key-one",
|
||||
sessionId: "session-independent-replacement",
|
||||
}),
|
||||
executePython("print('two')", {
|
||||
cwd: "/tmp/replacement-key-two",
|
||||
sessionId: "session-independent-replacement",
|
||||
}),
|
||||
]);
|
||||
|
||||
expect(startCount).toBe(4);
|
||||
expect(replacementOne.executeCalls).toEqual(["print('one')"]);
|
||||
expect(replacementTwo.executeCalls).toEqual(["print('two')"]);
|
||||
});
|
||||
|
||||
it("resets the session kernel when requested", async () => {
|
||||
const firstKernel = new FakeKernel(okResult);
|
||||
const secondKernel = new FakeKernel(okResult);
|
||||
|
||||
Reference in New Issue
Block a user