99.99% Uptime SLAArchitecture & Telemetry Case Study

Microservices API Gateway

High-throughput API routing engine built with Node.js, Express, and Redis token-bucket rate-limiting algorithm to protect downstream backend microservices from traffic surges.

Proxmox VE Node Telemetry & Telemetry Monitor

Node: gateway-sg-01.arhab.net

Status: Online • Uptime: 7d 12:32:00 • Kernel: Linux 6.12.101-edge (32 GB RAM)

Cluster: sg-production-01

Throughput

1420 req/s

Cache Hit Ratio

94.2%

P99 Latency

18 ms

429 Blocks/min

12 req

vCPU Core UtilizationsAvg Load: 28%

Core 0[|||......................] 12%

Core 1[||.......................] 8%

Core 2[|||||||||||..............] 45%

Core 3[||||||...................] 23%

Core 4[||||.....................] 14%

Core 5[|........................] 5%

Core 6[||||||||||||||||||||||...] 89%

Core 7[||||||||.................] 31%

System Resources & I/O

RAM Usage12.42 GB / 32.00 GB

Redis Cache RAM128.4 MB / 1.00 GB

Network RX199.1 MB/s

Network TX233.4 MB/s

session: gateway-sg-01 [htop-monitor]

TASKS: 48 total, 1 running

1 [|||......................] 12%

2 [||.......................] 8%

3 [|||||||||||..............] 45%

4 [||||||...................] 23%

5 [||||.....................] 14%

6 [|........................] 5%

Uptime: 7d 12:32:00

Load average: 0.42 0.38 0.31

PIDUSERPRINIVIRTRESSHRSCPU%MEM%TIME+Command

1402gateway2001.2G142M45MS4.2%3.5%12h:04:12node gateway/server.js

1408redis200512M128M12MS2.1%3.2%14d:06:12redis-server *:6379

1412envoy200890M64M28MS1.8%1.6%14d:06:12envoy -c gateway.yaml

1420nginx200120M12M8MS0.8%0.3%14d:06:12nginx: worker process

1422root20085M8M4MS0.1%0.2%0h:00:15htop

F1HelpF2SetupF3SearchF4FilterF5TreeF6SortByF9KillF10Quit

v3.3.0

Problem & Challenge

Uncontrolled traffic bursts on public endpoints were causing database thread pool exhaustion and cascading HTTP 504 timeouts across dependent downstream microservices.

Engineering Solution

Deployed a stateless API gateway layer backed by atomic Redis Lua scripts implementing a sliding-window token bucket algorithm. Excess requests are instantly rejected at the edge with HTTP 429 without hitting core application servers.