What is in scope
- Envelope ingest only
- 70/30 small and medium error mix
- Issue and event query probes after load
Evaluate urgentry
Read the numbers with the caveats attached.
These benchmarks compare Tiny mode, urgentry self-hosted, and Sentry self-hosted on the same host with the same narrow error-tracking workload. The goal is not theater. The goal is to help you choose the right operator shape.
400 eps
Stable Tiny throughput on the current same-host benchmark lane.
2,200 eps
Stable self-hosted throughput on the same comparison host.
2.2x more
Same-host self-hosted throughput than the current Sentry reference row.
48.82 ms
Current self-hosted query p95 on the large-box reference lane.
How to read this
The workload is narrow on purpose.
These comparisons focus on envelope ingest, a 70/30 small and medium error mix, and the same issue and event query probes after load.
What this page is for
- Picking Tiny vs self-hosted
- Comparing memory and latency tradeoffs
- Reading the current Sentry same-host reference row honestly
Published line
These numbers are tied to a dated artifact trail.
The current public line is anchored to the April 13, 2026 reruns. The large-box host for that published comparison was urgentry-test on Ubuntu 24.04 with 8 vCPU, 30.6 GiB RAM, and a 226 GB root disk.
Artifact trail
- Published reference date: 2026-04-13
- Tiny benchmark artifact
- Self-hosted benchmark artifact
- Feature-matrix artifact
How to cite it safely
- Same-host comparison, not a universal benchmark claim
- Error-tracking workload only, not a whole observability stack bake-off
- Use the artifact links when you need the exact current public numbers
Large-box reference
The current main comparison line.
| Target | Stable load | Ingest p95 | Query p95 | Peak memory |
|---|---|---|---|---|
| Tiny | 400 eps | 10.08 ms | 78.66 ms | 52.3 MB |
| Self-hosted | 2,200 eps | 0.71 ms | 48.82 ms | 391.8 MB |
| Sentry self-hosted | 1,000 eps | 0.62 ms | 1400.81 ms | 8191.7 MB |
- Tiny is the easiest first stop and stays in a low-memory class.
- Self-hosted buys more headroom and better query latency, but it costs more operationally.
- Sentry still wins raw ingest latency on this host, but uses far more memory and much slower post-load queries.
Feature tradeoffs
What the self-hosted build tags change.
| Variant | Tags | Stable load | Query p95 | Peak memory |
|---|---|---|---|---|
| Default self-hosted | default |
2200 eps | 50.48 ms | 392 MB |
| Controller build tag | k8scontroller |
2200 eps | 47.68 ms | 414.4 MB |
| ClickHouse pilot | clickhouse |
2500 eps | 44.71 ms | 442.7 MB |
| ClickHouse + controller | clickhouse,k8scontroller |
2500 eps | 47.58 ms | 427 MB |
- The binary-size and memory tradeoff are the clearest signals.
- The spread between 2200 and 2500 eps should be treated as directional on this shared host.
- The feature matrix is there to separate real regressions from host noise.
Small-box reference
Use this when you are sizing a cheap VPS.
| Target | Result | Validation note |
|---|---|---|
| Tiny | 100 eps; ingest p95 6.8 ms; query p95 79.9 ms; peak RSS 44.8 MB | core ingest and query flow worked |
| Self-hosted | 100 eps; ingest p95 5.3 ms; query p95 55.5 ms; peak stack RSS 184.0 MB | basic split-role flow worked |
| Sentry self-hosted | install did not complete | the official install stalled the Docker control plane on this host |
This smaller-box run is useful when you want to know whether Tiny is enough for a cheap VPS and whether self-hosted is worth the extra memory and patience.
Next docs
Use the migration and self-hosted docs when you are ready to connect the numbers back to an actual rollout decision.