Unhandled errors in ANR worker causes app to pause indefinitely

Is there an existing issue for this?

I have checked for existing issues https://github.com/getsentry/sentry-javascript/issues
I have reviewed the documentation https://docs.sentry.io/
I am using the latest SDK release https://github.com/getsentry/sentry-javascript/releases

How do you use Sentry?

Sentry Saas (sentry.io)

Which SDK are you using?

@sentry/node
@sentry/integrations

SDK Version

7.107.0 (48b8169)

Framework Version

No response

Link to Sentry event

No response

SDK Setup

const Sentry = require('@sentry/node') as typeof import('@sentry/node')
const Integrations = require('@sentry/integrations') as typeof import('@sentry/integrations')
// SENTRY_RELEASE is a special object that must be defined on `global` and will be picked up by
// the sentry SDK to report the release associated with the event. We use compile-time variable
// substitution in the bundler to replace the RELEASE_CODE constant with its real value. This
// can also be "injected" by the esbuild-sentry plugin but this adds function calls to every
// single bundled module.
global.SENTRY_RELEASE = { id: RELEASE_CODE }
Sentry.init({
	dsn: "https://...@....ingest.sentry.io/...",
	attachStacktrace: true,
	includeLocalVariables: true,
	sendDefaultPii: true,
	maxBreadcrumbs: 40,
	normalizeDepth: 6,
	integrations: [
		new Integrations.CaptureConsole({
			levels: ['warn', 'error']
		}),
		new Sentry.Integrations.Anr({
			captureStackTrace: true,
			anrThreshold: 50,
			pollInterval: 50,
		})
	]
})

Sentry.setUser({
	id: hostname
})

Sentry.setContext('device', {
	hostname: hostname
})

Steps to Reproduce

Make sure there is a const ctx defined in the main node runtime
Enable ANR worker
Cause ANR events (If ctx not defined in your own code, then causing more than 1 anr event I believe also triggers the issue because then ctx has already been defined)
Notice that the program just stays hanging:

This is due to multiple issues with the worker.ts script:

sentry-javascript/packages/node/src/integrations/anr/worker.ts

Lines 171 to 192 in fa1621c

    
           session.post( 
        
             'Runtime.evaluate', 
        
             { 
        
               // Grab the trace context from the current scope 
        
               expression: 
        
                 'const ctx = __SENTRY__.acs?.getCurrentScope().getPropagationContext() || {}; ctx.traceId + "-" + ctx.spanId + "-" + ctx.parentSpanId', 
        
               // Don't re-trigger the debugger if this causes an error 
        
               silent: true, 
        
             }, 
        
             (_, param) => { 
        
               const traceId = param && param.result ? (param.result.value as string) : '--'; 
        
               const [trace_id, span_id, parent_span_id] = traceId.split('-') as (string | undefined)[]; 
        
               session.post('Debugger.resume'); 
        
               session.post('Debugger.disable'); 
        
               const context = trace_id?.length && span_id?.length ? { trace_id, span_id, parent_span_id } : undefined; 
        
               sendAnrEvent(stackFrames, context).then(null, () => { 
        
                 log('Sending ANR event failed.'); 
        
               }); 
        
             }, 
        
           );

Errors in evaluation of the const ctx = __SENTRY__... expression designed to capture the data for the trace ID are not properly handled.

For example if ctx is already defined, the expression throws Identifier 'ctx' has already been declared. Here is the value of param coming into the callback when that occurs:

const thing = {
	result: {
		type: 'object',
		subtype: 'error',
		className: 'SyntaxError',
		description: "SyntaxError: Identifier 'ctx' has already been declared\n" +
			'    at eval (eval at Codegen (/.../index.js:33122:29), <anonymous>:1:1)\n' +
			'    at Function (<anonymous>)\n' +
			'    at Codegen (/.../index.js:33122:29)\n' +
			'    at Type.setup (/.../index.js:34876:51)\n' +
			'    at Type.fromObject (/.../index.js:34912:19)\n' +
			'    at Type.Write$fromObject [as fromObject] (eval at Codegen (/.../index.js:33122:29), <anonymous>:35:34)\n' +
			'    at Type.fromObject (/.../index.js:34912:27)\n' +
			'    at Type.WriteRequest$fromObject [as fromObject] (eval at Codegen (/.../index.js:33122:29), <anonymous>:20:24)\n' +
			'    at Object.serialize [as requestSerialize] (/.../index.js:39147:29)\n' +
			'    at BaseStreamingInterceptingCall.sendMessageWithContext (/.../index.js:30755:46)',02: 21: 46.156 AM[0ms]
		objectId: '2536922713100491570.1.278'
	},
	exceptionDetails: {
		exceptionId: 2,
		text: 'Uncaught',
		lineNumber: 0,
		columnNumber: 0,
		scriptId: '594',
		exception: {02: 21: 46.156 AM[0ms]
			type: 'object',
			subtype: 'error',
			className: 'SyntaxError',02: 21: 46.157 AM[1ms]
			description: "SyntaxError: Identifier 'ctx' has already been declared\n" +
				'    at eval (eval at Codegen (/.../index.js:33122:29), <anonymous>:1:1)\n' +02: 21: 46.157 AM [0ms]
			'    at Function (<anonymous>)\n' +
				'    at Codegen (/.../index.js:33122:29)\n' +
				'    at Type.setup (/.../index.js:34876:51)\n' +
				'    at Type.fromObject (/.../index.js:34912:19)\n' +
				'    at Type.Write$fromObject [as fromObject] (eval at Codegen (/.../index.js:33122:29), <02:21:46.157 AM [   0ms] anonymous>:35:34)\n' +
				'    at Type.fromObject (/.../index.js:34912:27)\n' +
				'    at Type.WriteRequest$fromObject [as fromObject] (eval at Codegen (/.../index.js:33122:29), <anonymous>:20:24)\n' +02: 21: 46.158 AM [1ms]
			'    at Object.serialize [as requestSerialize] (/.../index.js:39147:29)\n' +
				'    at BaseStreamingInterceptingCall.sendMessageWithContext (/.../index.js:30755:46)',
			objectId: '2536922713100491570.1.279'
		}
	}
}

When an error like this occurs in evaluating the expression, this leads to TypeError: Cannot read properties of undefined (reading 'split') since traceId is not actually a string but rather undefined

This error gets silently dropped, and the program remains paused and stuck, since the following Debugger.resume never gets called

Expected Result

The ANR event should have been created and reported, and my process should have been resumed and continued running as normal.

The Debugger.resume should have been sent back to the runtime FIRST to make sure that it always resumes, and:

any error that occured in evaluating the const ctx = __SENTRY__.hub.getScope... expression and
any error that occurs in the callback of session.post('Runtime.evaluate'

Should be at least surfaced with a console.warn or something.

Actual Result

With captureStackTrace disabled, I am also observing that in sendAnrEvent the worker thread is exited and is never restarted, so we only ever get one Anr event from a single instance of the app?

sentry-javascript/packages/node/src/integrations/anr/worker.ts

Line 133 in fa1621c

process.exit(0);

I also only get user details (id etc.) on the AppNotResponding event if I modify the event inside the worker to add user

This applies with or without captureStackTrace enabled

Example of a regular event where user ID, app memory, start time etc. all captured normally: 3eb53f70c4974c6bb21f41e46c5c1f47
Example of an ANR event where nothing else was captured: 1ba31386009441fa8f5f16b2b975e4af

Both of these were triggered with the same setup - just a setTimeout which either threw an Error or sat in a blocking loop for 2s to trigger the ANR.

Another one: sendAnrEvent doesn't seem to pass through tunnel to createEventEnvelope, which now explains we are not getting ANR events from devices out on site, even after patching this other stuff - without tunnel being honoured our ANR event requests would be getting dropped by firewall.

@delaneyb

@delaneyb thanks for the detailed report! We will take a look and hopefully address al your feedback.

@timfish Would you mind taking a look when you find the time? I think the ctx collision should be pretty straightforward by just renaming to something a little bit less prone to colliding. The other feedback is a bit trickier to implement I think and might even be additional features on top of our current ANR implementation.

I've split this into multiple issues:

ctx variable (this issue)
tunnel option ANR worker does not respect tunnel option #11159
Multiple events ANR only sends one event #11160
Scope sync (already open issue) Include more scope context for ANR events #10668

fix(node): Use unique variable for ANR context transfer (#11161)

CTX variable released with https://github.com/getsentry/sentry-javascript/releases/tag/7.108.0

fix(node): Use unique variable for ANR context transfer (getsentry#11161

Navigation Menu

Repository navigation

Unhandled errors in ANR worker causes app to pause indefinitely #11147

Is there an existing issue for this?

How do you use Sentry?

Which SDK are you using?

SDK Version

Framework Version

Link to Sentry event

SDK Setup

Steps to Reproduce

Expected Result

Actual Result

Metadata

Assignees

Labels

Projects

Milestone

Relationships

Development

	session.post(
	'Runtime.evaluate',
	{
	// Grab the trace context from the current scope
	expression:
	'const ctx = __SENTRY__.acs?.getCurrentScope().getPropagationContext() \|\| {}; ctx.traceId + "-" + ctx.spanId + "-" + ctx.parentSpanId',
	// Don't re-trigger the debugger if this causes an error
	silent: true,
	},
	(_, param) => {
	const traceId = param && param.result ? (param.result.value as string) : '--';
	const [trace_id, span_id, parent_span_id] = traceId.split('-') as (string \| undefined)[];

	session.post('Debugger.resume');
	session.post('Debugger.disable');

	const context = trace_id?.length && span_id?.length ? { trace_id, span_id, parent_span_id } : undefined;
	sendAnrEvent(stackFrames, context).then(null, () => {
	log('Sending ANR event failed.');
	});
	},
	);

Unhandled errors in ANR worker causes app to pause indefinitely #11147

Description

Is there an existing issue for this?

How do you use Sentry?

Which SDK are you using?

SDK Version

Framework Version

Link to Sentry event

SDK Setup

Steps to Reproduce

Expected Result

Actual Result

Activity

delaneyb commented on Mar 17, 2024

delaneyb commented on Mar 17, 2024

delaneyb commented on Mar 17, 2024

lforst commented on Mar 18, 2024

timfish commented on Mar 18, 2024

AbhiPrasad commented on Mar 22, 2024

Metadata

Metadata

Assignees

Labels

Projects

Milestone

Relationships

Development

Issue actions