GitRiver Load Benchmark

Every number here comes from a run on the hardware listed below, not from an estimate: the harness brings up an installation, seeds it with the data listed below and drives the scenarios to saturation.

Test environment

Parameter	Value
GitRiver version	1.1.0
Run date	2026-09-17
CPU	Intel(R) Xeon(R) CPU E5-2667 v4 @ 3.20GHz (16 threads)
RAM	94 GB
Storage	NVMe SSD, btrfs
Kernel	Linux 7.2.4
PostgreSQL	18.6
Build profile	release
Run conditions	dedicated machine, all other load stopped
Server binary size	57.2 MB

Seed data:

Entity	Count
Users	1 000
Repositories	10 000
Issues	100 000
Activity records	100 000
Pull requests	100 000
Reviews	200 000
Review requests	100 000
Pull request assignees	100 000
Repository directory (on disk)	1.45 GB
Database	247.2 MB

The load client and the server run on the same machine: the network is excluded from the measurement, but they share CPU time. The request rate limit is lifted for the duration of the run — otherwise the write scenarios would measure the rate limiter, not the server.

Memory at idle

The server runs with the seeded database and its background jobs active, serving no requests. After startup there is a 30-second wait, then a 60-second measurement window; the table records the peak over that window, not a lucky instant.

Metric	Value
Resident memory at idle	45.8 MB
Binary size	57.2 MB

Scenario summary

For each scenario the table takes the step with the highest throughput among the runs that produced no failures at all — that is, the regime before the breaking point, not past it. The full step series are below.

Scenario	Call	Conc.	req/s	p50, ms	p95, ms	p99, ms	Errors	Peak memory
Repository explore	GET /api/v1/explore	128	9821.75	12.87	16.76	18.9	0	71.1 MB
Repository page	GET /api/v1/repos/{owner}/{name}	128	17554.9	7.23	10.18	11.86	0	71.5 MB
Issue list	GET /api/v1/repos/{owner}/{name}/issues	64	12657.5	4.91	7.54	9.31	0	67.2 MB
Repository pull requests	GET /api/v1/repos/{owner}/{name}/pull_requests	64	7602.8	8.29	10.51	11.93	0	280.9 MB
Activity feed	GET /api/v1/activity	16	654.7	24.15	27.59	29.75	0	202.7 MB
Dashboard summary	GET /api/v1/dashboard/stats	64	6719.45	9.42	11.26	12.33	0	300.4 MB
My pull requests	GET /api/v1/dashboard/prs	64	5876.8	10.68	14.24	16.2	0	261.1 MB
Pull requests to review	GET /api/v1/dashboard/review-prs	64	2664.35	23.64	30.17	33.9	0	225.2 MB
Code search	GET /api/v1/search/code	8	54.2	145.16	177.7	195.68	0	187.4 MB
Issue creation	POST /api/v1/repos/{owner}/{name}/issues	64	3136.55	16.2	59.41	105.9	0	318.9 MB
Clone (git clone)	git clone http://…/{owner}/{name}	16	123.85	129.08	146.56	156.25	0	51.0 MB
Push (git push)	git push http://…/{owner}/{name}	16	120.35	131.72	164.75	180.27	0	208.1 MB

Single-client response

The bottom step of every series: one client, no queue. This is the cost of the call itself — what a person waits for when opening a page on an idle node.

Each scenario is preceded by a separate warm-up pass whose result is discarded: otherwise the bottom step would measure the first read from disk rather than the server.

Scenario	p50, ms	p95, ms	p99, ms	req/s
Repository explore	2.48	2.68	3.09	418.55
Repository page	1.22	1.4	1.51	838.2
Issue list	1.86	2.24	2.37	546.55
Repository pull requests	2.79	3.16	3.36	366.5
Activity feed	22.05	24.28	25.28	45.4
Dashboard summary	2.43	2.8	2.97	422.55
My pull requests	3.61	3.91	4.46	287.9
Pull requests to review	5.95	7.33	8	169.4
Code search	25.58	27.56	28.97	38.95
Issue creation	2.79	4.12	4.81	319.35
Clone (git clone)	43.16	47.58	50.5	23.35
Push (git push)	58.56	95.36	105.07	13.85

Saturation behaviour

Every scenario is run at several concurrency levels. While throughput keeps growing the server is not saturated; the knee of the curve together with a rising p99 marks the saturation point. Both axes are logarithmic.

Repository explore

Conc.	req/s	p50, ms	p95, ms	p99, ms	max, ms	Errors	CPU, % of a core	Peak memory
1	418.55	2.48	2.68	3.09	5	0	39.7	43.0 MB
4	1431.9	2.81	3.69	4.13	10.54	0	145.95	46.7 MB
16	8391.4	1.84	2.69	3.14	10.13	0	480.55	58.0 MB
64	9812.5	6.34	9.86	11.74	21	0	580.4	64.6 MB
128	9821.75	12.87	16.76	18.9	29.73	0	587.55	71.1 MB

Repository page

Conc.	req/s	p50, ms	p95, ms	p99, ms	max, ms	Errors	CPU, % of a core	Peak memory
1	838.2	1.22	1.4	1.51	4.88	0	58.9	44.8 MB
4	2745.35	1.45	1.97	2.2	7.16	0	230.95	47.2 MB
16	14554.65	1.06	1.53	1.84	5.63	0	695.8	61.1 MB
64	17247.5	3.65	5.44	6.42	14.7	0	870.9	66.7 MB
128	17554.9	7.23	10.18	11.86	22.66	0	899.7	71.5 MB

Issue list

Conc.	req/s	p50, ms	p95, ms	p99, ms	max, ms	Errors	CPU, % of a core	Peak memory
1	546.55	1.86	2.24	2.37	4.22	0	57	45.2 MB
4	1909.85	2.09	2.83	3.2	13.95	0	215.95	47.4 MB
16	11096.6	1.39	1.98	2.37	9.03	0	672.4	61.4 MB
64	12657.5	4.91	7.54	9.31	18.43	0	808.9	67.2 MB
128	12408.8	9.93	15.2	19.46	27.94	0	808.7	73.4 MB

Repository pull requests

Conc.	req/s	p50, ms	p95, ms	p99, ms	max, ms	Errors	CPU, % of a core	Peak memory
1	366.5	2.79	3.16	3.36	9.71	0	63.1	45.9 MB
4	1457.05	2.78	3.6	3.97	7.18	0	241.8	104.4 MB
16	6916.15	2.26	3.02	3.46	6.78	0	563.65	205.6 MB
64	7602.8	8.29	10.51	11.93	19.29	0	660.3	280.9 MB
128	7503.45	16.75	20.53	23.93	35.38	0	661.95	303.0 MB

Activity feed

Conc.	req/s	p50, ms	p95, ms	p99, ms	max, ms	Errors	CPU, % of a core	Peak memory
1	45.4	22.05	24.28	25.28	26.42	0	14.05	44.9 MB
4	248.2	16.07	17.22	17.73	19.38	0	44.1	66.9 MB
16	654.7	24.15	27.59	29.75	38.53	0	148.75	202.7 MB
64	620.4	103.22	130.94	143.66	171.09	0	158.9	228.6 MB
128	617.55	208.46	264.79	287.71	335.49	0	167.55	264.8 MB

Dashboard summary

Conc.	req/s	p50, ms	p95, ms	p99, ms	max, ms	Errors	CPU, % of a core	Peak memory
1	422.55	2.43	2.8	2.97	7.3	0	60	44.8 MB
4	2316.1	1.66	2.4	2.76	4.23	0	230.75	104.6 MB
16	6438.4	2.43	3.32	3.84	8.34	0	428.4	222.5 MB
64	6719.45	9.42	11.26	12.33	16.88	0	467.6	300.4 MB
128	6636.15	19.16	21.36	22.71	28.45	0	471.4	264.8 MB

My pull requests

Conc.	req/s	p50, ms	p95, ms	p99, ms	max, ms	Errors	CPU, % of a core	Peak memory
1	287.9	3.61	3.91	4.46	15.88	0	45.25	44.7 MB
4	996.75	4.06	5.28	5.86	7.45	0	163.6	103.1 MB
16	5311.2	2.9	4.23	4.96	23.87	0	432.85	204.3 MB
64	5876.8	10.68	14.24	16.2	24.55	0	497.7	261.1 MB
128	5779.9	21.85	26.16	28.98	40.43	0	498.45	283.8 MB

Pull requests to review

Conc.	req/s	p50, ms	p95, ms	p99, ms	max, ms	Errors	CPU, % of a core	Peak memory
1	169.4	5.95	7.33	8	19.11	0	34.2	44.6 MB
4	752.95	5.25	7.6	8.63	14.24	0	129.85	103.8 MB
16	2525.05	6.15	9.06	10.6	20.43	0	257.25	204.6 MB
64	2664.35	23.64	30.17	33.9	44.15	0	283.7	225.2 MB
128	2640.3	47.87	57.09	62.89	79.92	0	285.15	300.1 MB

Code search

Conc.	req/s	p50, ms	p95, ms	p99, ms	max, ms	Errors	CPU, % of a core	Peak memory
1	38.95	25.58	27.56	28.97	30.3	0	51.3	57.3 MB
2	48.65	41.14	44.1	45.89	48.8	0	65	58.9 MB
4	53.25	74.86	83.79	88.77	94.01	0	71.7	78.5 MB
8	54.2	145.16	177.7	195.68	234.33	0	71.6	187.4 MB

Issue creation

Conc.	req/s	p50, ms	p95, ms	p99, ms	max, ms	Errors	CPU, % of a core	Peak memory
1	319.35	2.79	4.12	4.81	224.48	0	56.7	49.2 MB
4	726.05	2.97	5.73	41.32	1263.54	0	102.75	112.8 MB
16	1504.6	5.01	39.07	48.51	1501.04	0	214	223.9 MB
64	3136.55	16.2	59.41	105.9	217.87	0	372.3	318.9 MB
128	902.4	132.84	236.31	1143.34	1232.26	0	172.25	319.5 MB

Clone (git clone)

Conc.	req/s	p50, ms	p95, ms	p99, ms	max, ms	Errors	CPU, % of a core	Peak memory
1	23.35	43.16	47.58	50.5	76.39	0	4.1	44.4 MB
4	116.6	34.16	36.15	37.71	47.55	0	14.05	46.5 MB
16	123.85	129.08	146.56	156.25	177.53	0	17.05	51.0 MB

Push (git push)

Conc.	req/s	p50, ms	p95, ms	p99, ms	max, ms	Errors	CPU, % of a core	Peak memory
1	13.85	58.56	95.36	105.07	1606.45	0	9.15	48.4 MB
4	73.15	47.86	74.64	397.85	604.15	0	45.5	107.9 MB
16	120.35	131.72	164.75	180.27	209.09	0	112.7	208.1 MB

What this means

Claim	Measured	Verdict
Memory at idle — under 50 MB	45.8 MB	confirmed
Binary size — under 60 MB	57.2 MB	confirmed

Under load memory reaches 320 MB — the ceiling across every step of every scenario, including code search and accepting a push, that is the worst case rather than the average.