databuff-proxy Dual-Write Stability

databuff-proxy · SkyWalking Agent dual-write · QPS=35 · report completed 2026-07-30

Migrating from SkyWalking to DataBuff can be a direct Agent retarget, or a databuff-proxy dual-write: point Agents once at the proxy, fan out to OAP and DataBuff, then leave SkyWalking after validation. This post is the dual-write load & stability report; steps live in From SkyWalking to DataBuff.

Test environment & write demo

- **Item** — Details - **Topology** — Agent (145) → proxy (120) → SkyWalking OAP (114) + DataBuff Ingest (113) - **Proxy** — databuff-proxy · listen :11800 · workers=8 · queue_size=4096 - **Write demo** — Microservice call chain: 17 business services, 2 replicas each - **Middleware** — Redis, MySQL, MongoDB, Kafka, etc. — 9 total - **Total pods** — 43 pods · write load QPS=35 - **Units** — Process cores (not host-wide %) · span/s ≠ segment/s

Business write TPS

35

Same load on both sides

DataBuff ingest

11.3k

span/s

Same-QPS baseline

SkyWalking ingest

5.81k

segment/s

Same-QPS baseline

Per-path forwards

5838

req/s · symmetric · err/drop = 0

Proxy usage

0.89

cores (under 1) · RSS ~47 MB

Proxy host

CPU

16 vCPU

Memory

64 GiB

- **Scenario** — Cores · RSS · Notes - **Sustained full load** — ~0.86 · ~44 MB · Under 1 core - **Overnight 10h+** — 0.89 · ~47 MB · Under 1 core · flat RSS

Figures are process cores, not host-wide CPU %.

~6-hour monitoring screenshots

Process CPU · ~5% ≈ 0.8 cores on 16-core host · flat

Process RSS · ~47–60 MB · no growth

Host RX · steady traffic

Host TX · steady traffic

CPU/RSS from process detail; network from host120 (no process NIC metrics).

Abnormal scenarios

- **Scenario** — How · Behavior · Result - **One backend down** — DataBuff port unavailable · Bad path circuits open; other path keeps writing; proxy stays up · OK - **Disable one write path** — Admin API toggle · That path stops; other unaffected; re-enable recovers · OK - **Sustained high load** — QPS=35 continuous · No drops / errors on either path · OK - **Long run** — 10h+ without stop · Still forwarding; RSS flat · OK - **Restart after config** — Restart proxy · Comes back and dual-writes again · OK

Conclusion

- **Check** — Result - **Resource stability (under 1 core · flat RSS)** — Pass - **Abnormal scenarios** — Pass

Ready for production dual-write. See docs for both migration options; write throughput: Write Performance Comparison.

GitHub: https://github.com/databufflabs/databuff-proxy · Docs: Proxy Dual-Write Stability · Migration: From SkyWalking