How does Go's garbage collector work, and what knobs do you have to tune it?
Go uses a concurrent tri-color mark-and-sweep GC that runs mostly alongside the program. GOGC controls heap growth trigger and GOMEMLIMIT (Go 1.19+) sets a soft memory ceiling.
- 1
Tri-color mark and sweep: objects are white (not visited), grey (reachable but children not scanned), or black (fully scanned)
- 2
Concurrent: GC mark phase runs alongside the program — very short stop-the-world pauses (<1ms)
- 3
Write barrier: ensures correctness when the mutator (your code) modifies pointers during GC
- 4
Sweep phase reclaims white objects after mark completes
- 5
GC is triggered when heap size doubles since last GC (controlled by GOGC)
- 1
Profile first with pprof heap profiles before tuning GC parameters
- 2
GOGC=off + GOMEMLIMIT: disable time-based GC, rely on memory limit for triggering — good for batch jobs
- 3
Reduce allocations using sync.Pool for frequently allocated objects
- 4
Avoid pointer-heavy data structures — GC must trace every pointer
- 5
Use GODEBUG=gctrace=1 to see GC pause times and heap sizes in logs
Scenario Questions
0-2 years experience
- 1You added an in‑memory cache to a Go microservice and notice the process memory keeps growing even after the cache is cleared. How would you use Go's GC tools to investigate and fix the issue?
- 2If you set GOGC=50 in a small command‑line tool, what effect does that have on its memory usage and pause behavior compared to the default setting?
2-5 years experience
- 1During a load test, your Go service started experiencing occasional 200 ms GC pauses. Walk me through how you would identify the root cause and which GC knobs you would adjust.
- 2You need to reduce latency spikes in a high‑throughput API written in Go. Explain how you would balance GOGC, GODEBUG=gctrace, and runtime/pprof to tune the collector.
5-8 years experience
- 1Our platform runs dozens of Go services on shared VMs with limited RAM. Design a strategy for setting GC parameters across services to maximize overall throughput while keeping tail latency low.
- 2Explain how the concurrent mark‑sweep collector works under the hood and discuss the implications of increasing GOMAXPROCS on GC pause times in a CPU‑bound workload.
8+ years experience
- 1Our organization is standardizing on Go for all backend services. Propose a company‑wide policy for GC tuning, including monitoring, default GOGC values, and guidelines for when teams should deviate.
- 2Describe how you would design a self‑healing system that automatically adjusts Go GC knobs in response to observed latency and memory pressure across a fleet of services.
Follow-up Questions
- What trade‑offs does the concurrent collector introduce for latency‑sensitive workloads?
- How do you decide which GC knob to adjust when you see high pause times?
- Can you share an incident where GC behavior caused a production issue and how you resolved it?