How to create an RXjs RetryWhen with delay and limit on tries

You need to apply the limit to the retry signal, for instance if you only wanted 10 retries:

loadSomething(): Observable<SomeInterface> {
  return this.http.get(this.someEndpoint, commonHttpHeaders())
    .retryWhen(errors => 
      // Time shift the retry
      errors.delay(500)
            // Only take 10 items
            .take(10)
            // Throw an exception to signal that the error needs to be propagated
            .concat(Rx.Observable.throw(new Error('Retry limit exceeded!'))
    );

EDIT

Some of the commenters asked how to make sure that the last error is the one that gets thrown. The answer is a bit less clean but no less powerful, just use one of the flattening map operators (concatMap, mergeMap, switchMap) to check which index you are at.

Note: Using the new RxJS 6 pipe syntax for future proofing (this is also available in later versions of RxJS 5).

loadSomething(): Observable<SomeInterface> {
  const retryPipeline = 
    // Still using retryWhen to handle errors
    retryWhen(errors => errors.pipe(
      // Use concat map to keep the errors in order and make sure they
      // aren't executed in parallel
      concatMap((e, i) => 
        // Executes a conditional Observable depending on the result
        // of the first argument
        iif(
          () => i > 10,
          // If the condition is true we throw the error (the last error)
          throwError(e),
          // Otherwise we pipe this back into our stream and delay the retry
          of(e).pipe(delay(500)) 
        )
      ) 
  ));

  return this.http.get(this.someEndpoint, commonHttpHeaders())
    // With the new syntax you can now share this pipeline between uses
    .pipe(retryPipeline)
}