Timed Out RPCs - Count and Diagnose Remote Call Failures

Question

## Problem

You are given a log of RPC (remote procedure call) records: each entry is `(call_id, service, start_time_ms, end_time_ms)`. A call is considered "timed out" if `end_time_ms - start_time_ms > timeout_ms`. Note: some calls may have no matching end entry (dropped response) -- treat those as timed out.

Given the list of start events, end events separately, and `timeout_ms`, compute:
1. Total timed-out call count
2. Per-service timed-out count, sorted by count descending

```python
def analyze_timeouts(
    starts: list[tuple[str, str, int]],  # (call_id, service, start_ms)
    ends: list[tuple[str, int]],          # (call_id, end_ms)
    timeout_ms: int
) -> tuple[int, list[tuple[str, int]]]:
```

## Example

```
starts = [("c1","auth",0),("c2","db",10),("c3","auth",20)]
ends   = [("c1",600),("c2",50)]
timeout_ms = 500
# c1: 600ms -> timeout; c2: 40ms -> ok; c3: no end -> timeout
Output: (2, [("auth", 2)])
```

## Follow-ups

1. How would you detect cascading failures (a timeout on service A causing timeouts on dependent service B)?
2. What if `call_id` is not unique -- how do you match starts to ends?
3. Compute the 99th-percentile latency per service for non-timed-out calls.

Timed Out RPCs - Count and Diagnose Remote Call Failures

Question Details

Full Details

Topics

More from Google