Description
When uploadPart
or uploadIncompletePart
throws (is rejected) entire server crashes. This happens often with Scaleway Object Storage as their service fails with error 500 with non-informative console message An error was encountered in a non-retryable streaming request.
from AWS library. After listening for unhandledRejection
event I got some more information.
Unhandled Rejection at: Promise {
<rejected> S3ServiceException [InternalError]: We encountered an internal error. Please try again.
at throwDefaultError (/app/node_modules/@smithy/smithy-client/dist-cjs/index.js:867:20)
at /app/node_modules/@smithy/smithy-client/dist-cjs/index.js:876:5
at de_CommandError (/app/node_modules/@aws-sdk/client-s3/dist-cjs/index.js:4965:14)
at process.processTicksAndRejections (node:internal/process/task_queues:95:5)
at async /app/node_modules/@smithy/middleware-serde/dist-cjs/index.js:35:20
at async /app/node_modules/@aws-sdk/middleware-sdk-s3/dist-cjs/index.js:483:18
at async /app/node_modules/@smithy/middleware-retry/dist-cjs/index.js:321:38
at async /app/node_modules/@aws-sdk/middleware-flexible-checksums/dist-cjs/index.js:315:18
at async /app/node_modules/@aws-sdk/middleware-sdk-s3/dist-cjs/index.js:109:22
at async /app/node_modules/@aws-sdk/middleware-sdk-s3/dist-cjs/index.js:136:14 {
'$fault': 'client',
'$metadata': {
httpStatusCode: 500,
requestId: 'txg5c09a7e2fbe14ba89375-0067aa0421',
extendedRequestId: 'txg5c09a7e2fbe14ba89375-0067aa0421',
cfId: undefined
},
Code: 'InternalError',
RequestId: 'txg5c09a7e2fbe14ba89375-0067aa0421',
HostId: 'txg5c09a7e2fbe14ba89375-0067aa0421',
Resource: '<s3-object-key>.bin'
},
I believe the problem is that when a deferred
promise is created it does not handle rejections. It is only inserted into a list.
At the end promises are aggregated with
Promise.all
and returned to the caller where an rejection handler is eventually added.tus-node-server/packages/server/src/server.ts
Line 188 in 81eb03a
During this period if any of the promises in a list are rejected, the rejection is not handled.
Steps to reproduce
Use S3 server that sometimes throws errors (e.g. Scaleway Object Storage) or manually throw error. I modified uploadPart(metadata, readStream, partNumber)
to randomly reject. Add provided snippet at the top of this function. This will result in server crash.
const errorProbability = Math.random()
if (errorProbability < 0.2) {
return new Promise((resolve, reject) => {
setTimeout(() => {
reject(new Error('test'));
}, 1000);
})
}
Expected behavior
Server should not crash. Failed part upload should end the current upload with HTTP 500 Internal Server Error status code.
Observation
Promises (rejections) returned by acquiredPermit?.release()
and permit?.release()
are also not handled and would cause server to crash.