Description
Currently, the @sentry/browser
SDK has no mechanism to automatically clone the Hub
(and, consequently, Scope
), unlike @sentry/node
that automatically clones a new hub per Node Domain -- and we use that to isolate state for Node server handling concurrent requests.
This means that, in particular for @sentry/browser
, any kind of async code may end up accidentally mixing up state unless hubs are manually cloned and used per unit of concurrency.
This issue tracks improving automatic hub/scope management/propagation in @sentry/browser
. As a possible solution we may use (an optional support for) Zone.js.
For example:
setTag('a', 'tag');
withScope((scope) => {
setTimeout(() => {
captureMessage('test'); // no tags
}, 0);
});
getCurrentHub().getScope().clear();
hub.withScope(async scope => {
console.log('scope1', scope);
scope.setTag('me', 'you');
hub.withScope(async scope => {
// will have tag me: you
console.log('scope2', scope);
await asyncOperation();
hub.withScope(async scope => {
// won't have tag me: you
console.log('scope3', scope);
await asyncOperation();
});
});
});
The problem is apparent when some code needs to get something from the current scope, for example here is our fetch
instrumentation:
sentry-javascript/packages/tracing/src/browser/request.ts
Lines 169 to 178 in 61eda62
sentry-javascript/packages/tracing/src/utils.ts
Lines 54 to 56 in 61eda62
Because fetch
is async, the "current scope" is not necessarily in the state that one would expect. In the example above, when we call getActiveTransaction()
, the current hub/scope may have changed and lost reference to the current transaction.
Metadata
Metadata
Assignees
Type
Projects
Status