Bump fulmine to 5.5.0 by nigrosimone · Pull Request #1085 · MDA2AV/HttpArena

nigrosimone · 2026-08-07T12:32:12Z

Pin bump for the fulmine entry, from ^5.2.0 to ^5.3.0 (changelog).

The release is mostly compatibility work: 27 fixes, nearly all of them found by a differential fuzzer that builds random applications and compares the answers against real Express. Nothing there should move a benchmark row.

Four changes do touch the request path, each measured on our own rig against the previous revision, so I have no prediction for yours:

the response arms its two internal listeners by writing the event map directly instead of calling on() twice, -6.9% CPU per request on a hello-world
the request no longer builds a Readable until something asks for one, -5.2% and -5.5% over two runs
an IPv4-mapped peer address is read straight out of the bytes rather than through the general IPv6 formatter, which shows on anything that reads req.ip
a chain steps over a body parser when the request declared no body and the verb reads none, which removes a hop rather than shortening one

No entry files change other than the pin.

nigrosimone · 2026-08-07T12:32:25Z

C:/Program Files/Git/benchmark -f fulmine --save

github-actions · 2026-08-07T12:32:33Z

👋 /benchmark request received. A collaborator will review and approve the run.

nigrosimone · 2026-08-07T12:36:10Z

Sorry, my shell rewrote the leading slash in the comment above into a Windows path. The command I meant:

/benchmark -f fulmine --save

github-actions · 2026-08-07T12:36:26Z

👋 /benchmark request received. A collaborator will review and approve the run.

…ers once

nigrosimone · 2026-08-07T12:59:46Z

Pushed a second commit, so this PR is now the pin bump plus a change to the crud cache. Happy to split it if you would rather review them apart.

ioredis → @redis/client, with the cache it can keep in the process. The client is created with RESP: 3 and clientSideCache: { ttl: 200, maxEntries: 4096, evictPolicy: 'LRU' }. A cache HIT is then answered inside the worker with no round trip, and Redis pushes the invalidation when a key stops being true, so the sidecar still owns correctness. The cache is bounded and LRU on purpose: there is one process per core, and 64 unbounded caches would be the working-set problem the sidecar was introduced to solve.

Two details that needed care:

connect() is a promise here, where ioredis connected in the background. Until it settles the in-process map answers, which is what already happened when the harness provided no sidecar: a cache that is not ready is a miss, never an error. Verified with a client pointed at a dead port: the fallback holds and nothing is left unhandled.
a write drops the key from this process too, and after the write rather than before it. The invalidation Redis pushes is a round trip away, and a read on the same connection immediately after the write must not beat it home.

enableAutoPipelining has no counterpart because node-redis has no such switch: it queues commands and flushes them on setImmediate, so pipelining is not optional there.

The crud read path built its headers per request. res.set({ ...SERVER_HDR, 'x-cache': 'HIT' }) is two allocations on the busiest route the entry has. They are two constants now.

Smoke tested in Docker against postgres 17 and redis 7 with the items table seeded: read MISS then HIT on the same connection, 404 for an id that is not there, list by category, POST upsert 201, and PUT followed by a read that comes back MISS with the new values, which is the invalidation doing its job. What I could not run here is scripts/validate.sh: it needs --network host, which Docker Desktop does not expose to a Windows host.

github-actions · 2026-08-07T13:45:52Z

Benchmark Results

Framework: fulmine | Test: all tests

Test	Conn	RPS	CPU	Mem	Δ RPS	Δ Mem
baseline	512	1,225,488	6815.6%	7.6GiB	~0%	+28.8%
baseline	4096	1,202,481	6777.0%	7.7GiB	-0.7%	+30.5%
pipelined	512	7,745,305	6571.4%	6.2GiB	+5.0%	+3.3%
pipelined	4096	7,720,147	6555.0%	6.8GiB	+5.2%	+4.6%
limited-conn	512	1,034,112	6140.0%	7.6GiB	~0%	+1.3%
limited-conn	4096	1,101,872	6637.7%	7.7GiB	+0.1%	+1.3%
json	4096	1,145,533	6636.1%	6.7GiB	+3.0%	+4.7%
json-comp	512	409,597	6243.9%	8.7GiB	-4.0%	+13.0%
json-comp	4096	417,359	6406.8%	9.2GiB	-2.6%	+1.1%
json-comp	16384	433,927	6320.4%	10.7GiB	-1.0%	+2.9%
json-tls	4096	1,085,015	6587.8%	6.5GiB	+3.7%	-1.5%
upload	32	2,038	3100.2%	6.5GiB	-0.1%	-4.4%
upload	256	2,130	6334.2%	7.0GiB	-0.3%	-1.4%
api-4	256	35,403	389.4%	508MiB	-1.0%	+1.6%
api-16	1024	129,162	1554.7%	1.9GiB	+4.5%	~0%
static	1024	538,479	6524.5%	7.2GiB	+0.4%	-1.4%
static	4096	573,270	6527.6%	7.9GiB	-1.1%	+6.8%
static	6800	564,169	6467.5%	8.2GiB	+1.0%	~0%
async-db	1024	236,060	5855.3%	7.0GiB	+4.0%	+1.4%
crud	4096	210,973	3236.6%	11.2GiB	-40.1%	+38.3%
echo-ws	512	3,707,047	6445.1%	2.3GiB	-0.7%	~0%
echo-ws	4096	3,768,441	6379.4%	2.4GiB	~0%	+4.3%
echo-ws	16384	3,586,702	6503.4%	2.5GiB	+0.5%	+4.2%
echo-ws-pipeline	512	24,603,148	6787.3%	2.3GiB	-1.3%	~0%
echo-ws-pipeline	4096	25,021,235	6631.2%	2.4GiB	+0.5%	~0%
echo-ws-pipeline	16384	23,550,083	6333.6%	2.6GiB	-0.8%	+4.0%
echo-ws-limited	512	2,082,245	5481.8%	2.3GiB	-0.7%	~0%
echo-ws-limited	4096	2,570,787	6246.0%	2.4GiB	+0.5%	~0%

Full log

  Bandwidth:  154.17MB/s
  WS upgrades: 16384
  WS frames:   115118784
  Latency samples: 115118784 / 115118784 responses (100.0%)
[info] CPU 6767.2% | Mem 2.6GiB

[run 3/3]
gcannon v0.5.3 [WS]
  Target:    localhost:8080/ws
  Threads:   64
  Conns:     16384 (256/thread)
  Pipeline:  16
  Req/conn:  unlimited (keep-alive)
  Expected:  200
  Duration:  5s


  Thread Stats   Avg      p50      p90      p99    p99.9
    Latency   11.45ms   11.10ms   13.20ms   17.30ms   21.70ms

  113793712 frames sent in 5.00s, 113531568 frames received
  Throughput: 22.70M req/s
  Bandwidth:  152.06MB/s
  WS upgrades: 16384
  WS frames:   113531568
  Latency samples: 113531568 / 113531568 responses (100.0%)
[info] CPU 6463.9% | Mem 2.6GiB

=== Best: 23550083 req/s (CPU: 6333.6%, Mem: 2.6GiB) ===
[info] saved results/echo-ws-pipeline/16384/fulmine.json
httparena-bench-fulmine
httparena-bench-fulmine

==============================================
=== fulmine / echo-ws-limited / 512c (tool=gcannon) ===
==============================================
[info] waiting for server...
[info] server ready

[run 1/3]
gcannon v0.5.3 [WS]
  Target:    localhost:8080/ws
  Threads:   64
  Conns:     512 (8/thread)
  Pipeline:  1
  Req/conn:  10
  Expected:  200
  Duration:  5s


  Thread Stats   Avg      p50      p90      p99    p99.9
    Latency    149us    120us    261us    561us   2.03ms

  10411181 frames sent in 5.00s, 10411225 frames received
  Throughput: 2.08M req/s
  Bandwidth:  46.85MB/s
  WS upgrades: 1056931
  WS frames:   10411225
  Latency samples: 10411189 / 10411225 responses (100.0%)
  Reconnects: 1041137
[info] CPU 5481.8% | Mem 2.3GiB

[run 2/3]
gcannon v0.5.3 [WS]
  Target:    localhost:8080/ws
  Threads:   64
  Conns:     512 (8/thread)
  Pipeline:  1
  Req/conn:  10
  Expected:  200
  Duration:  5s


  Thread Stats   Avg      p50      p90      p99    p99.9
    Latency    143us    118us    257us    527us   1.12ms

  10345404 frames sent in 5.00s, 10345523 frames received
  Throughput: 2.07M req/s
  Bandwidth:  46.56MB/s
  WS upgrades: 1049740
  WS frames:   10345523
  Latency samples: 10345481 / 10345523 responses (100.0%)
  Reconnects: 1034607
[info] CPU 5572.5% | Mem 2.3GiB

[run 3/3]
gcannon v0.5.3 [WS]
  Target:    localhost:8080/ws
  Threads:   64
  Conns:     512 (8/thread)
  Pipeline:  1
  Req/conn:  10
  Expected:  200
  Duration:  5s


  Thread Stats   Avg      p50      p90      p99    p99.9
    Latency    143us    118us    256us    517us   1.08ms

  10387526 frames sent in 5.00s, 10387625 frames received
  Throughput: 2.08M req/s
  Bandwidth:  46.75MB/s
  WS upgrades: 1054677
  WS frames:   10387625
  Latency samples: 10387581 / 10387625 responses (100.0%)
  Reconnects: 1038808
[info] CPU 5512.3% | Mem 2.4GiB

=== Best: 2082245 req/s (CPU: 5481.8%, Mem: 2.3GiB) ===
[info] saved results/echo-ws-limited/512/fulmine.json
httparena-bench-fulmine
httparena-bench-fulmine

==============================================
=== fulmine / echo-ws-limited / 4096c (tool=gcannon) ===
==============================================
[info] waiting for server...
[info] server ready

[run 1/3]
gcannon v0.5.3 [WS]
  Target:    localhost:8080/ws
  Threads:   64
  Conns:     4096 (64/thread)
  Pipeline:  1
  Req/conn:  10
  Expected:  200
  Duration:  5s


  Thread Stats   Avg      p50      p90      p99    p99.9
    Latency   1.05ms    866us   1.77ms   3.17ms   5.99ms

  12811584 frames sent in 5.00s, 12809146 frames received
  Throughput: 2.56M req/s
  Bandwidth:  57.69MB/s
  WS upgrades: 1285355
  WS frames:   12809146
  Latency samples: 12809137 / 12809146 responses (100.0%)
  Reconnects: 1279825
[info] CPU 5976.6% | Mem 2.4GiB

[run 2/3]
gcannon v0.5.3 [WS]
  Target:    localhost:8080/ws
  Threads:   64
  Conns:     4096 (64/thread)
  Pipeline:  1
  Req/conn:  10
  Expected:  200
  Duration:  5s


  Thread Stats   Avg      p50      p90      p99    p99.9
    Latency   1.06ms    930us   1.79ms   2.95ms   3.94ms

  12856556 frames sent in 5.00s, 12853935 frames received
  Throughput: 2.57M req/s
  Bandwidth:  57.92MB/s
  WS upgrades: 1287779
  WS frames:   12853935
  Latency samples: 12857001 / 12853935 responses (100.0%)
  Reconnects: 1284262
[info] CPU 6246.0% | Mem 2.4GiB

[run 3/3]
gcannon v0.5.3 [WS]
  Target:    localhost:8080/ws
  Threads:   64
  Conns:     4096 (64/thread)
  Pipeline:  1
  Req/conn:  10
  Expected:  200
  Duration:  5s


  Thread Stats   Avg      p50      p90      p99    p99.9
    Latency   1.07ms    953us   1.80ms   2.96ms   3.89ms

  12845714 frames sent in 5.00s, 12845453 frames received
  Throughput: 2.57M req/s
  Bandwidth:  57.83MB/s
  WS upgrades: 1291236
  WS frames:   12845453
  Latency samples: 12845416 / 12845453 responses (100.0%)
  Reconnects: 1284409
[info] CPU 6060.9% | Mem 2.4GiB

=== Best: 2570787 req/s (CPU: 6246.0%, Mem: 2.4GiB) ===
[info] saved results/echo-ws-limited/4096/fulmine.json
httparena-bench-fulmine
httparena-bench-fulmine
[info] rebuilding site/data/*.json
[updated] /home/diogo/actions-runner/_work/HttpArena/HttpArena/site/data/frameworks.json
[updated] /home/diogo/actions-runner/_work/HttpArena/HttpArena/site/data/results/fulmine.json - 28 new, 28 total
[updated] /home/diogo/actions-runner/_work/HttpArena/HttpArena/site/data/current.json
[info] done
httparena-postgres
httparena-redis
[info] restoring loopback MTU to 65536

…hput

nigrosimone · 2026-08-07T14:02:30Z

Thanks for running it. One row in that result is mine and it is a regression, so I have taken the change back.

crud at 4096 connections, against the previous run of this entry:

	before	after
rps	352,481	210,973
avg latency	11.56ms	19.41ms
p99	21.20ms	43.90ms
cpu	4502%	3236%
memory	8.1GiB	11.2GiB
non-2xx	0	0

The CPU is the interesting column: it went down while latency doubled, and nothing failed, so the
processes were waiting rather than working. That is a bottleneck that moved out of this entry and into
Redis. Server-assisted client-side caching makes Redis keep an invalidation table per key per client
and push a message to all 64 of them whenever one writes, and this test writes on every cache miss.

The last commit puts ioredis back exactly as it was, with the reason written next to it so nobody
tries it again. Two things from those commits are kept because they are independent of the client:
the two crud response-header objects are built once instead of being spread per request, and the body
reader joins its chunks as buffers so a UTF-8 character split across two chunks survives.

Sorry for the wasted run. Whenever you can re-run it, crud should be back around 352k; the rest of
the table was already fine, with pipelined, json, json-tls, api-16 and async-db up 3-5%
from the 5.3.0 bump itself.

nigrosimone · 2026-08-07T14:09:51Z

/benchmark -f lightning --save

github-actions · 2026-08-07T14:10:02Z

👋 /benchmark request received. A collaborator will review and approve the run.

nigrosimone · 2026-08-07T16:35:12Z

Bumped to 5.4.0, which is on npm now, so this PR is one thing to approve rather than two: the
ioredis revert and the new version get measured together.

What is in it that this benchmark can see:

crud should return to around 352k. That row is the reason for the revert, and the previous run's
40% drop is explained in the comment above.
Four compatibility fixes, none of which change the shapes this benchmark exercises: a mount no
longer answers past a layer written before it, res.format's 406 leaves the route as Express's
does, req.query hands back a fresh object per read as Express's getter does, and
res.getHeader("content-length") answers instead of coming back empty.
Two small ones on the response path: it no longer builds its Writable state until something asks,
measured at 45ns a response, and the peer-address read no longer reopens its window every 100000
requests.

Nothing here should move the other rows outside noise. If any of them does, I would rather hear it
than not, since this rig is steadier than mine.

nigrosimone · 2026-08-07T16:44:35Z

/benchmark -f lightning --save

github-actions · 2026-08-07T16:44:42Z

👋 /benchmark request received. A collaborator will review and approve the run.

MDA2AV · 2026-08-07T17:35:42Z

/benchmark -f fulmine --save

github-actions · 2026-08-07T17:35:50Z

👋 /benchmark request received. A collaborator will review and approve the run.

MDA2AV · 2026-08-07T17:36:05Z

there is no "lightning" framework, I think it somehow translated from italian to english lol

github-actions · 2026-08-07T17:52:03Z

Benchmark Results

Framework: fulmine | Test: all tests

Test	Conn	RPS	CPU	Mem	Δ RPS	Δ Mem
baseline	512	1,128,115	6762.7%	2.6GiB	-8.0%	-55.9%
baseline	4096	1,106,235	6588.8%	2.6GiB	-8.6%	-55.9%
pipelined	512	7,896,176	6579.7%	2.5GiB	+7.0%	-58.3%
pipelined	4096	7,882,640	6564.7%	2.6GiB	+7.5%	-60.0%
limited-conn	512	957,714	6215.6%	2.6GiB	-7.3%	-65.3%
limited-conn	4096	1,017,734	6450.8%	2.7GiB	-7.5%	-64.5%
json	4096	1,061,146	6655.1%	2.6GiB	-4.6%	-59.4%
json-comp	512	400,205	6223.2%	8.5GiB	-6.2%	+10.4%
json-comp	4096	411,258	6400.2%	9.0GiB	-4.0%	-1.1%
json-comp	16384	423,202	6330.7%	10.5GiB	-3.5%	+1.0%
json-tls	4096	994,024	6601.1%	2.6GiB	-5.0%	-60.6%
upload	32	2,040	3058.9%	6.7GiB	~0%	-1.5%
upload	256	2,140	6313.2%	7.1GiB	+0.1%	~0%
api-4	256	34,674	393.6%	403MiB	-3.0%	-19.4%
api-16	1024	126,921	1556.1%	1.2GiB	+2.7%	-36.8%
static	1024	541,909	6513.4%	3.7GiB	+1.0%	-49.3%
static	4096	585,570	6512.9%	7.4GiB	+1.0%	~0%
static	6800	570,414	6500.2%	7.7GiB	+2.2%	-6.1%
async-db	1024	228,797	5870.9%	4.1GiB	+0.8%	-40.6%
crud	4096	357,052	4458.0%	8.0GiB	+1.3%	-1.2%
echo-ws	512	3,713,621	6436.3%	2.3GiB	-0.5%	~0%
echo-ws	4096	3,781,001	6545.8%	2.4GiB	+0.3%	+4.3%
echo-ws	16384	3,587,532	6497.5%	2.5GiB	+0.5%	+4.2%
echo-ws-pipeline	512	24,733,760	6791.5%	2.3GiB	-0.8%	~0%
echo-ws-pipeline	4096	25,011,145	6603.9%	2.4GiB	+0.4%	~0%
echo-ws-pipeline	16384	23,840,864	6349.7%	2.6GiB	+0.4%	+4.0%
echo-ws-limited	512	2,086,949	5568.7%	2.4GiB	-0.5%	+4.3%
echo-ws-limited	4096	2,567,800	6250.3%	2.4GiB	+0.4%	~0%

Full log

  Bandwidth:  155.21MB/s
  WS upgrades: 16384
  WS frames:   115888304
  Latency samples: 115888304 / 115888304 responses (100.0%)
[info] CPU 6768.1% | Mem 2.6GiB

[run 3/3]
gcannon v0.5.3 [WS]
  Target:    localhost:8080/ws
  Threads:   64
  Conns:     16384 (256/thread)
  Pipeline:  16
  Req/conn:  unlimited (keep-alive)
  Expected:  200
  Duration:  5s


  Thread Stats   Avg      p50      p90      p99    p99.9
    Latency   11.40ms   10.90ms   13.30ms   17.60ms   22.10ms

  114610912 frames sent in 5.00s, 114352864 frames received
  Throughput: 22.86M req/s
  Bandwidth:  153.15MB/s
  WS upgrades: 16384
  WS frames:   114352864
  Latency samples: 114352864 / 114352864 responses (100.0%)
[info] CPU 6499.1% | Mem 2.6GiB

=== Best: 23840864 req/s (CPU: 6349.7%, Mem: 2.6GiB) ===
[info] saved results/echo-ws-pipeline/16384/fulmine.json
httparena-bench-fulmine
httparena-bench-fulmine

==============================================
=== fulmine / echo-ws-limited / 512c (tool=gcannon) ===
==============================================
[info] waiting for server...
[info] server ready

[run 1/3]
gcannon v0.5.3 [WS]
  Target:    localhost:8080/ws
  Threads:   64
  Conns:     512 (8/thread)
  Pipeline:  1
  Req/conn:  10
  Expected:  200
  Duration:  5s


  Thread Stats   Avg      p50      p90      p99    p99.9
    Latency    145us    117us    255us    547us   2.21ms

  10283973 frames sent in 5.00s, 10284020 frames received
  Throughput: 2.06M req/s
  Bandwidth:  46.28MB/s
  WS upgrades: 1044043
  WS frames:   10284020
  Latency samples: 10285603 / 10284020 responses (100.0%)
  Reconnects: 1028405
[info] CPU 5464.4% | Mem 2.3GiB

[run 2/3]
gcannon v0.5.3 [WS]
  Target:    localhost:8080/ws
  Threads:   64
  Conns:     512 (8/thread)
  Pipeline:  1
  Req/conn:  10
  Expected:  200
  Duration:  5s


  Thread Stats   Avg      p50      p90      p99    p99.9
    Latency    147us    121us    260us    536us   1.35ms

  10434688 frames sent in 5.00s, 10434748 frames received
  Throughput: 2.09M req/s
  Bandwidth:  46.96MB/s
  WS upgrades: 1058358
  WS frames:   10434748
  Latency samples: 10434713 / 10434748 responses (100.0%)
  Reconnects: 1043488
[info] CPU 5568.7% | Mem 2.4GiB

[run 3/3]
gcannon v0.5.3 [WS]
  Target:    localhost:8080/ws
  Threads:   64
  Conns:     512 (8/thread)
  Pipeline:  1
  Req/conn:  10
  Expected:  200
  Duration:  5s


  Thread Stats   Avg      p50      p90      p99    p99.9
    Latency    145us    120us    259us    528us   1.17ms

  10404195 frames sent in 5.00s, 10404277 frames received
  Throughput: 2.08M req/s
  Bandwidth:  46.82MB/s
  WS upgrades: 1054908
  WS frames:   10404277
  Latency samples: 10404243 / 10404277 responses (100.0%)
  Reconnects: 1040468
[info] CPU 5478.6% | Mem 2.4GiB

=== Best: 2086949 req/s (CPU: 5568.7%, Mem: 2.4GiB) ===
[info] saved results/echo-ws-limited/512/fulmine.json
httparena-bench-fulmine
httparena-bench-fulmine

==============================================
=== fulmine / echo-ws-limited / 4096c (tool=gcannon) ===
==============================================
[info] waiting for server...
[info] server ready

[run 1/3]
gcannon v0.5.3 [WS]
  Target:    localhost:8080/ws
  Threads:   64
  Conns:     4096 (64/thread)
  Pipeline:  1
  Req/conn:  10
  Expected:  200
  Duration:  5s


  Thread Stats   Avg      p50      p90      p99    p99.9
    Latency   1.07ms    881us   1.78ms   3.21ms   6.12ms

  12802653 frames sent in 5.00s, 12801086 frames received
  Throughput: 2.56M req/s
  Bandwidth:  57.65MB/s
  WS upgrades: 1285123
  WS frames:   12801086
  Latency samples: 12801066 / 12801086 responses (100.0%)
  Reconnects: 1279236
[info] CPU 5970.4% | Mem 2.4GiB

[run 2/3]
gcannon v0.5.3 [WS]
  Target:    localhost:8080/ws
  Threads:   64
  Conns:     4096 (64/thread)
  Pipeline:  1
  Req/conn:  10
  Expected:  200
  Duration:  5s


  Thread Stats   Avg      p50      p90      p99    p99.9
    Latency   1.05ms    906us   1.76ms   2.90ms   3.86ms

  12837912 frames sent in 5.00s, 12839000 frames received
  Throughput: 2.57M req/s
  Bandwidth:  57.78MB/s
  WS upgrades: 1287878
  WS frames:   12839000
  Latency samples: 12838928 / 12839000 responses (100.0%)
  Reconnects: 1284380
[info] CPU 6250.3% | Mem 2.4GiB

[run 3/3]
gcannon v0.5.3 [WS]
  Target:    localhost:8080/ws
  Threads:   64
  Conns:     4096 (64/thread)
  Pipeline:  1
  Req/conn:  10
  Expected:  200
  Duration:  5s


  Thread Stats   Avg      p50      p90      p99    p99.9
    Latency   1.06ms    934us   1.80ms   2.97ms   4.69ms

  12827468 frames sent in 5.00s, 12827893 frames received
  Throughput: 2.56M req/s
  Bandwidth:  57.74MB/s
  WS upgrades: 1292103
  WS frames:   12827893
  Latency samples: 12827845 / 12827893 responses (100.0%)
  Reconnects: 1282824
[info] CPU 6049.3% | Mem 2.4GiB

=== Best: 2567800 req/s (CPU: 6250.3%, Mem: 2.4GiB) ===
[info] saved results/echo-ws-limited/4096/fulmine.json
httparena-bench-fulmine
httparena-bench-fulmine
[info] rebuilding site/data/*.json
[updated] /home/diogo/actions-runner/_work/HttpArena/HttpArena/site/data/frameworks.json
[updated] /home/diogo/actions-runner/_work/HttpArena/HttpArena/site/data/results/fulmine.json - 28 new, 28 total
[updated] /home/diogo/actions-runner/_work/HttpArena/HttpArena/site/data/current.json
[info] done
httparena-postgres
httparena-redis
[info] restoring loopback MTU to 65536

nigrosimone · 2026-08-07T18:22:30Z

Fulmine = Lightning
:D

nigrosimone · 2026-08-08T05:00:01Z

Bumped to 5.4.1. The last run's -4..-9% on the query-carrying rows was real, it was mine, and this release removes it.

Expected on re-run: baseline, limited-conn, json, json-tls, json-comp and api-4 back to their 5.3.0 numbers, crud staying at ~357k, everything else unchanged.

nigrosimone · 2026-08-08T05:02:35Z

/benchmark -f fulmine --save

github-actions · 2026-08-08T05:02:42Z

👋 Benchmark request received. A collaborator will review and approve the run.

github-actions · 2026-08-08T05:03:04Z

⚠️ /benchmark --save cannot start: main has diverged and cannot be auto-merged into this branch. Please merge or rebase main manually, push, and re-run /benchmark --save.

nigrosimone · 2026-08-08T05:41:25Z

Merged upstream/main in, so the branch is mergeable again and /benchmark --save should start. The three commits it picked up (static-tls, the pipelined reference-only change, the new PR commands) touch nothing under frameworks/fulmine, and the entry is unchanged at ^5.4.1.

Same expectation as before: crud around 357k, and the query-carrying rows (baseline, limited-conn, json, json-tls, json-comp, api-4) back to their 5.3.0 numbers now that the copy-per-read is gone.

github-actions · 2026-08-08T05:41:34Z

👋 Benchmark request received. A collaborator will review and approve the run.

nigrosimone · 2026-08-08T06:21:15Z

Bumped again, to 5.5.0, which is on npm. Nothing in it changes what this benchmark exercises: it adds res.flushHeaders() and fixes two TypeScript declarations, all three found by running an Angular SSR application on it. The performance-relevant change is still the one from 5.4.1, the req.query fix.

Expectation for the run is unchanged: crud around 357k, and baseline, limited-conn, json, json-tls, json-comp and api-4 back to their 5.3.0 numbers.

The branch is mergeable with main as of the merge two commits back.

github-actions · 2026-08-08T08:38:23Z

Benchmark Results

Framework: fulmine | Test: all tests

Test	Conn	RPS	CPU	Mem	Δ RPS	Δ Mem
baseline	512	1,230,527	6816.6%	2.6GiB	+0.4%	-55.9%
baseline	4096	1,212,009	6776.3%	2.6GiB	+0.1%	-55.9%
pipelined	512	7,896,553	6584.6%	2.5GiB	+7.0%	-58.3%
pipelined	4096	7,864,864	6569.7%	2.6GiB	+7.2%	-60.0%
limited-conn	512	1,032,024	6147.7%	2.6GiB	-0.1%	-65.3%
limited-conn	4096	1,102,548	6412.2%	2.7GiB	+0.2%	-64.5%
json	4096	1,141,097	6649.7%	2.6GiB	+2.6%	-59.4%
json-comp	512	412,239	6227.9%	8.7GiB	-3.4%	+13.0%
json-comp	4096	425,219	6419.6%	9.2GiB	-0.8%	+1.1%
json-comp	16384	438,127	6312.7%	10.6GiB	~0%	+1.9%
json-tls	4096	1,076,647	6586.0%	2.6GiB	+2.9%	-60.6%
upload	32	2,024	2997.7%	6.6GiB	-0.8%	-2.9%
upload	256	2,129	6107.8%	7.1GiB	-0.4%	~0%
api-4	256	35,639	393.3%	425MiB	-0.3%	-15.0%
api-16	1024	129,658	1542.9%	1.2GiB	+4.9%	-36.8%
static	1024	543,570	6509.7%	3.9GiB	+1.3%	-46.6%
static	4096	588,146	6545.6%	7.3GiB	+1.4%	-1.4%
static	6800	567,798	6487.7%	7.6GiB	+1.7%	-7.3%
async-db	1024	224,285	5942.1%	3.6GiB	-1.2%	-47.8%
crud	4096	357,460	4430.6%	8.0GiB	+1.4%	-1.2%
echo-ws	512	3,726,934	6467.6%	2.3GiB	-0.1%	~0%
echo-ws	4096	3,768,590	6547.5%	2.4GiB	~0%	+4.3%
echo-ws	16384	3,577,210	6232.0%	2.5GiB	+0.2%	+4.2%
echo-ws-pipeline	512	24,949,417	6817.8%	2.3GiB	~0%	~0%
echo-ws-pipeline	4096	24,928,886	6589.2%	2.4GiB	+0.1%	~0%
echo-ws-pipeline	16384	23,866,108	6339.2%	2.6GiB	+0.5%	+4.0%
echo-ws-limited	512	2,089,072	5617.6%	2.3GiB	-0.3%	~0%
echo-ws-limited	4096	2,564,958	6043.9%	2.4GiB	+0.3%	~0%

Full log

  Bandwidth:  155.22MB/s
  WS upgrades: 16384
  WS frames:   115897872
  Latency samples: 115897872 / 115897872 responses (100.0%)
[info] CPU 6774.7% | Mem 2.6GiB

[run 3/3]
gcannon v0.5.3 [WS]
  Target:    localhost:8080/ws
  Threads:   64
  Conns:     16384 (256/thread)
  Pipeline:  16
  Req/conn:  unlimited (keep-alive)
  Expected:  200
  Duration:  5s


  Thread Stats   Avg      p50      p90      p99    p99.9
    Latency   11.37ms   10.90ms   13.30ms   17.30ms   22.00ms

  114962448 frames sent in 5.00s, 114700304 frames received
  Throughput: 22.93M req/s
  Bandwidth:  153.62MB/s
  WS upgrades: 16384
  WS frames:   114700304
  Latency samples: 114700304 / 114700304 responses (100.0%)
[info] CPU 6491.1% | Mem 2.6GiB

=== Best: 23866108 req/s (CPU: 6339.2%, Mem: 2.6GiB) ===
[info] saved results/echo-ws-pipeline/16384/fulmine.json
httparena-bench-fulmine
httparena-bench-fulmine

==============================================
=== fulmine / echo-ws-limited / 512c (tool=gcannon) ===
==============================================
[info] waiting for server...
[info] server ready

[run 1/3]
gcannon v0.5.3 [WS]
  Target:    localhost:8080/ws
  Threads:   64
  Conns:     512 (8/thread)
  Pipeline:  1
  Req/conn:  10
  Expected:  200
  Duration:  5s


  Thread Stats   Avg      p50      p90      p99    p99.9
    Latency    148us    120us    260us    547us   1.89ms

  10387264 frames sent in 5.00s, 10387343 frames received
  Throughput: 2.08M req/s
  Bandwidth:  46.75MB/s
  WS upgrades: 1053414
  WS frames:   10387343
  Latency samples: 10387286 / 10387343 responses (100.0%)
  Reconnects: 1038759
[info] CPU 5493.9% | Mem 2.3GiB

[run 2/3]
gcannon v0.5.3 [WS]
  Target:    localhost:8080/ws
  Threads:   64
  Conns:     512 (8/thread)
  Pipeline:  1
  Req/conn:  10
  Expected:  200
  Duration:  5s


  Thread Stats   Avg      p50      p90      p99    p99.9
    Latency    147us    120us    259us    531us   1.52ms

  10445283 frames sent in 5.00s, 10445364 frames received
  Throughput: 2.09M req/s
  Bandwidth:  47.01MB/s
  WS upgrades: 1057189
  WS frames:   10445364
  Latency samples: 10445329 / 10445364 responses (100.0%)
  Reconnects: 1044563
[info] CPU 5617.6% | Mem 2.3GiB

[run 3/3]
gcannon v0.5.3 [WS]
  Target:    localhost:8080/ws
  Threads:   64
  Conns:     512 (8/thread)
  Pipeline:  1
  Req/conn:  10
  Expected:  200
  Duration:  5s


  Thread Stats   Avg      p50      p90      p99    p99.9
    Latency    144us    119us    256us    526us   1.37ms

  10368939 frames sent in 5.00s, 10369015 frames received
  Throughput: 2.07M req/s
  Bandwidth:  46.66MB/s
  WS upgrades: 1052581
  WS frames:   10369015
  Latency samples: 10368981 / 10369015 responses (100.0%)
  Reconnects: 1036941
[info] CPU 5493.0% | Mem 2.4GiB

=== Best: 2089072 req/s (CPU: 5617.6%, Mem: 2.3GiB) ===
[info] saved results/echo-ws-limited/512/fulmine.json
httparena-bench-fulmine
httparena-bench-fulmine

==============================================
=== fulmine / echo-ws-limited / 4096c (tool=gcannon) ===
==============================================
[info] waiting for server...
[info] server ready

[run 1/3]
gcannon v0.5.3 [WS]
  Target:    localhost:8080/ws
  Threads:   64
  Conns:     4096 (64/thread)
  Pipeline:  1
  Req/conn:  10
  Expected:  200
  Duration:  5s


  Thread Stats   Avg      p50      p90      p99    p99.9
    Latency   1.06ms    884us   1.76ms   3.13ms   6.13ms

  12786832 frames sent in 5.00s, 12786917 frames received
  Throughput: 2.56M req/s
  Bandwidth:  57.54MB/s
  WS upgrades: 1284015
  WS frames:   12786917
  Latency samples: 12786888 / 12786917 responses (100.0%)
  Reconnects: 1278009
[info] CPU 6001.3% | Mem 2.4GiB

[run 2/3]
gcannon v0.5.3 [WS]
  Target:    localhost:8080/ws
  Threads:   64
  Conns:     4096 (64/thread)
  Pipeline:  1
  Req/conn:  10
  Expected:  200
  Duration:  5s


  Thread Stats   Avg      p50      p90      p99    p99.9
    Latency   1.05ms    947us   1.72ms   2.82ms   3.73ms

  12814731 frames sent in 5.00s, 12814431 frames received
  Throughput: 2.56M req/s
  Bandwidth:  57.70MB/s
  WS upgrades: 1286396
  WS frames:   12814431
  Latency samples: 12814402 / 12814431 responses (100.0%)
  Reconnects: 1281611
[info] CPU 6236.5% | Mem 2.4GiB

[run 3/3]
gcannon v0.5.3 [WS]
  Target:    localhost:8080/ws
  Threads:   64
  Conns:     4096 (64/thread)
  Pipeline:  1
  Req/conn:  10
  Expected:  200
  Duration:  5s


  Thread Stats   Avg      p50      p90      p99    p99.9
    Latency   1.06ms    929us   1.78ms   2.85ms   3.78ms

  12826266 frames sent in 5.00s, 12824793 frames received
  Throughput: 2.56M req/s
  Bandwidth:  57.75MB/s
  WS upgrades: 1287619
  WS frames:   12824793
  Latency samples: 12824791 / 12824793 responses (100.0%)
  Reconnects: 1281697
[info] CPU 6043.9% | Mem 2.4GiB

=== Best: 2564958 req/s (CPU: 6043.9%, Mem: 2.4GiB) ===
[info] saved results/echo-ws-limited/4096/fulmine.json
httparena-bench-fulmine
httparena-bench-fulmine
[info] rebuilding site/data/*.json
[updated] /home/diogo/actions-runner/_work/HttpArena/HttpArena/site/data/frameworks.json
[updated] /home/diogo/actions-runner/_work/HttpArena/HttpArena/site/data/results/fulmine.json - 28 new, 28 total
[updated] /home/diogo/actions-runner/_work/HttpArena/HttpArena/site/data/current.json
[info] done
httparena-postgres
httparena-redis
[info] restoring loopback MTU to 65536

nigrosimone · 2026-08-08T09:16:13Z

It's good for me! 🎉

Bump fulmine to 5.3.0

0de2163

Use @redis/client with its client-side cache, and build the crud head…

7d035ca

…ers once

nigrosimone requested review from Kaliumhexacyanoferrat and MDA2AV as code owners August 7, 2026 12:59

nigrosimone and others added 2 commits August 7, 2026 15:06

Join the body chunks as buffers, so a split character survives

8651672

Benchmark results: fulmine [skip ci]

5ab7831

Go back to ioredis: the client-side cache cost crud 40% of its throug…

475ed4e

…hput

Bump fulmine to 5.4.0

416855a

nigrosimone changed the title ~~Bump fulmine to 5.3.0~~ Bump fulmine to 5.4.0 Aug 7, 2026

Benchmark results: fulmine [skip ci]

890f114

MDA2AV approved these changes Aug 7, 2026

View reviewed changes

Bump fulmine to 5.4.1

c1865c9

nigrosimone changed the title ~~Bump fulmine to 5.4.0~~ Bump fulmine to 5.4.1 Aug 8, 2026

Merge upstream/main into bump-fulmine-530

2226ae8

Bump fulmine to 5.5.0

d829078

nigrosimone changed the title ~~Bump fulmine to 5.4.1~~ Bump fulmine to 5.5.0 Aug 8, 2026

Benchmark results: fulmine [skip ci]

8b24bb3

MDA2AV merged commit 8e85bf7 into MDA2AV:main Aug 8, 2026

Uh oh!

Conversation

nigrosimone commented Aug 7, 2026

Uh oh!

nigrosimone commented Aug 7, 2026

Uh oh!

github-actions Bot commented Aug 7, 2026

Uh oh!

nigrosimone commented Aug 7, 2026

Uh oh!

github-actions Bot commented Aug 7, 2026

Uh oh!

nigrosimone commented Aug 7, 2026

Uh oh!

github-actions Bot commented Aug 7, 2026

Benchmark Results

Uh oh!

nigrosimone commented Aug 7, 2026

Uh oh!

nigrosimone commented Aug 7, 2026

Uh oh!

github-actions Bot commented Aug 7, 2026

Uh oh!

nigrosimone commented Aug 7, 2026

Uh oh!

nigrosimone commented Aug 7, 2026

Uh oh!

github-actions Bot commented Aug 7, 2026

Uh oh!

MDA2AV commented Aug 7, 2026

Uh oh!

github-actions Bot commented Aug 7, 2026

Uh oh!

MDA2AV commented Aug 7, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

github-actions Bot commented Aug 7, 2026

Benchmark Results

Uh oh!

nigrosimone commented Aug 7, 2026

Uh oh!

nigrosimone commented Aug 8, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

nigrosimone commented Aug 8, 2026

Uh oh!

github-actions Bot commented Aug 8, 2026

Uh oh!

github-actions Bot commented Aug 8, 2026

Uh oh!

nigrosimone commented Aug 8, 2026

Uh oh!

github-actions Bot commented Aug 8, 2026

Uh oh!

nigrosimone commented Aug 8, 2026

Uh oh!

github-actions Bot commented Aug 8, 2026

Benchmark Results

Uh oh!

nigrosimone commented Aug 8, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

MDA2AV commented Aug 7, 2026 •

edited

Loading

nigrosimone commented Aug 8, 2026 •

edited

Loading