Long thumbnail generation time for S3 loader

[blitss]

I'm testing imagorvideo along with s3 storage loader and generating a thumbnail (uncached) consistently takes a very long time, more than 20 seconds every time.

And that is when running on beefy dedicated Hetzner server which I use for encoding the videos.
Is there any way to optimize that time and is that expected? I'm running pretty much the default settings and only set seek to 10 seconds, e.g.:

http://localhost:8000/unsafe/1920x1080/filters:seek(10s)/sl/obj_01j6fxp6ggf2pbarp9276sc4fa.mp4

downloading the same file itself in full is taking only 5 seconds:

$ time curl --output /dev/null
% Total % Received % Xferd Average Speed Time Time Time Current
Dload Upload Total Spent Left Speed
100 232M 100 232M 0 0 42.8M 0 0:00:05 0:00:05 --:--:-- 45.0M
curl --output /dev/null 0.31s user 0.50s system 14% cpu 5.431 total

so I guess a lot of time is spent on generating the thumbnail, what can I do to reduce that? the video info:

Metadata:
major_brand : isom
minor_version : 512
compatible_brands: isomiso2avc1mp41
creation_time : 2024-08-25T16:17:54.000000Z
encoder : Blackmagic Design DaVinci Resolve Studio
Duration: 00:00:58.11, start: 0.000000, bitrate: 33523 kb/s
Stream #0:00x1: Video: h264 (High) (avc1 / 0x31637661), yuv420p(tv, bt709, progressive), 3840x2160 [SAR 1:1 DAR 16:9], 33333 kb/s, 23.98 fps, 23.98 tbr, 24k tbn (default)

I'm also using avif + webp auto settings but that seem to have no impact for performance

[cshum]

Would you please try this docker build, which I am testing imagor with seekable s3 read stream.
See if it makes any difference:
https://github.com/cshum/imagorvideo/pkgs/container/imagorvideo/266301862?tag=master

ghcr.io/cshum/imagor@sha256:31a24a7cc2eba7efbbc59050358932b4654a22d763237228fe8a0fdcf356d634

[blitss]

Would you please try this docker build, which I am testing imagor with seekable s3 read stream. See if it makes any difference: https://github.com/cshum/imagorvideo/pkgs/container/imagorvideo/266301862?tag=master

ghcr.io/cshum/imagor@sha256:31a24a7cc2eba7efbbc59050358932b4654a22d763237228fe8a0fdcf356d634

Doesn't make any difference, it's still above 20s
For experiment I tried using frame(n) instead of automatic frame selection and that does seem to make no difference either

Used this image: shumc/imagorvideo@sha256:31a24a7cc2eba7efbbc59050358932b4654a22d763237228fe8a0fdcf356d634

[blitss]

Ok it seems like I ran into some nested virtualization issues here and the problem is actually on my side performance-wise.
Because I ran the same container on ARM Mac and even with Rosetta intel code it runs faster lol.

Sorry about that.

[blitss]

hm no actually I think I was right the first time the s3 loader time is higher than http

Compared using imagor@s3, imagor@http and just ffmpeg with the same input
using cloudflare r2 storage here

Method	Time (seconds)
FFmpeg	1.654
Imagor via HTTP loader	0.981
Imagor via S3 loader	20.310

Using ffmpeg

root@ffmpeg-pod:/# time ffmpeg -y -i https://r2v.**.app/obj_01j6m0ncdpe35bhyybgzzv259x.mp4 -ss 00:00:01 -frames:v 1 -vf scale=1920:1080 -update 1 -f image2 output_thumbnail.jpg
ffmpeg version n7.0.2 Copyright (c) 2000-2024 the FFmpeg developers
built with gcc 13 (Ubuntu 13.2.0-23ubuntu4)
configuration: --prefix=/root/ffmpeg-build --cc=gcc --ld=g++ --cxx=g++ --pkg-config-flags=--static --extra-cflags=-I/root/ffmpeg-build/include --extra-ldflags='-L/root/ffmpeg-build/lib -L/root/ffmpeg-build/lib64' --extra-libs='-lpthread -lm -lz' --extra-ldexeflags=-static --bindir=/root/bin --enable-gpl --disable-shared --enable-static --enable-nonfree --enable-openssl --disable-libfreetype --enable-libfdk-aac --enable-libmp3lame --enable-libx264 --enable-libx265 --enable-libaom --enable-libsvtav1 --enable-libvmaf --enable-version3 --enable-libopus --disable-ffplay --disable-doc --disable-podpages --disable-debug
libavutil 59. 8.100 / 59. 8.100
libavcodec 61. 3.100 / 61. 3.100
libavformat 61. 1.100 / 61. 1.100
libavdevice 61. 1.100 / 61. 1.100
libavfilter 10. 1.100 / 10. 1.100
libswscale 8. 1.100 / 8. 1.100
libswresample 5. 1.100 / 5. 1.100
libpostproc 58. 1.100 / 58. 1.100
Input #0, mov,mp4,m4a,3gp,3g2,mj2, from 'https://r2v.**.app/obj_01j6m0ncdpe35bhyybgzzv259x.mp4':
Metadata:
major_brand : mp42
minor_version : 0
compatible_brands: mp42mp41isomavc1
creation_time : 2021-06-11T18:48:27.000000Z
Duration: 00:00:15.96, start: 0.000000, bitrate: 22420 kb/s
Stream #0:00x1: Video: h264 (High) (avc1 / 0x31637661), yuv420p(tv, bt709, progressive), 4096x2160, 22417 kb/s, 25 fps, 25 tbr, 25 tbn (default)
Metadata:
creation_time : 2021-06-11T18:48:27.000000Z
handler_name : L-SMASH Video Handler
vendor_id : [0][0][0][0]
encoder : AVC Coding
Stream mapping:
Stream #0:0 -> #0:0 (h264 (native) -> mjpeg (native))
Press [q] to stop, [?] for help
[swscaler @ 0x7ff774012080] deprecated pixel format used, make sure you did set range correctly
Output #0, image2, to 'output_thumbnail.jpg':
Metadata:
major_brand : mp42
minor_version : 0
compatible_brands: mp42mp41isomavc1
encoder : Lavf61.1.100
Stream #0:0(und): Video: mjpeg, yuvj420p(pc, unknown/bt709/bt709, progressive), 1920x1080, q=2-31, 200 kb/s, 25 fps, 25 tbn (default)
Metadata:
creation_time : 2021-06-11T18:48:27.000000Z
handler_name : L-SMASH Video Handler
vendor_id : [0][0][0][0]
encoder : Lavc61.3.100 mjpeg
Side data:
cpb: bitrate max/min/avg: 0/0/200000 buffer size: 0 vbv_delay: N/A
[out#0/image2 @ 0x65482c0] video:81KiB audio:0KiB subtitle:0KiB other streams:0KiB global headers:0KiB muxing overhead: unknown
frame= 1 fps=0.8 q=5.3 Lsize=N/A time=00:00:00.04 bitrate=N/A speed=0.0315x

real 0m1.654s
user 0m2.428s
sys 0m0.177s

Using imagor via HTTP loader

$ time curl --output thumbnail.png "https://imagor.**.app/unsafe/1920x1080/filters:frame(2)/https://r2v.**.app/obj_01j6m0ncdpe35bhyybgzzv259x.
mp4"
% Total % Received % Xferd Average Speed Time Time Time Current
Dload Upload Total Spent Left Speed
100 138k 100 138k 0 0 143k 0 --:--:-- --:--:-- --:--:-- 143k

real 0m0.981s
user 0m0.013s
sys 0m0.012s

Using imagor via S3 loader

$ time curl --output thumbnail.png "https://imagor.**.app/unsafe/1920x1080/filters:frame(1s)/sl/obj_01j6m0ncdpe35bhyybgzzv259x.mp4"
% Total % Received % Xferd Average Speed Time Time Time Current
Dload Upload Total Spent Left Speed
100 136k 100 136k 0 0 6885 0 0:00:20 0:00:20 --:--:-- 34300

real 0m20.310s
user 0m0.016s
sys 0m0.014s

[cshum]

Can you share any of the mp4 files to help reproduce the issue?

[blitss]

Can you share any of the mp4 files to help reproduce the issue?

Sure
https://r2v.streamloop.app/obj_01j6m0ncdpe35bhyybgzzv259x.mp4

It does seem tho that problem is only reproducible by loading from S3 storage (I'm using CloudFlare, but also reproducible with others)