0

I have got Amazon S3 bucket with some strange behavior. When a to a request similar to GET /?delimiter=%2F&marker=&max-keys=1000&prefix= to get contents or a "root" folder of a bucket, i am getting success headers back, but a response body is delayed. Depending on a timeout i am settings , it can receive different parts of a body. It looks like i am getting < symbol, after a second ? and later x, m, l, and so on. Data is received with some delays almost after every byte . If i set a request timeout to 100 seconds i almost always have full response. If i set timeout to 30 seconds i can get empty body or something like "

This strange problem happens only in one S3 account and with one bucket in that account. All other bucket are fine.

Same behavior i see when use the tool s3cli . I used the https traffic sniffer i i really see how bytes appears with a delay from a network .

BUT. When i use the too Cyberduck for Amazon S3, it works fast with same bucket!

I tried to experiment with http headers, make them same as cyberduck, but no success. For my curl based tool problem exists

My request is as this one

GET /?delimiter=%2F&marker=&max-keys=1000&prefix= HTTP/1.1
User-Agent: My S3 Explorer
Host: s4-******.s3.amazonaws.com
x-amz-request-payer: requester
X-Amz-Date: 20190920T151135Z
Date: Fri, 20 Sep 2019 15:11:35 GMT
x-amz-content-sha256: e3******55
Authorization: AWS4-HMAC-SHA256 *****
Connection: Keep-Alive

And request from Cyberduck is as this

CONNECT s4-*********.s3.amazonaws.com:443 HTTP/1.1
Host: *********.s3.amazonaws.com:443
User-Agent: Cyberduck/7.0.1.30930 (Mac OS X/10.14) (x86_64)
HTTP/1.0 200 Connection established
GET /?max-keys=1000&versions&prefix&delimiter=%2F HTTP/1.1
Date: Fri, 20 Sep 2019 09:26:20 GMT
x-amz-request-payer: requester
x-amz-content-sha256: e3*********55
Host: s4-********.s3.amazonaws.com
x-amz-date: 20190920T092620Z
Authorization: ********
Connection: Keep-Alive
User-Agent: Cyberduck/7.0.1.30930 (Mac OS X/10.14) (x86_64)

It returns full response fast. But why?

There is one difference, before doing a request it executed some CONNECT command (i see this from sniffer) . Can that command affect? If yes, how do repeat same with curl (php)?

What can be the secret here?

Roman Gelembjuk
  • 1,797
  • 2
  • 25
  • 50
  • Have you tried to use curl from bash, from different locations? Experiment with different version of HTTP protocol, max-keys limit. Maybe resolve of host is slow? I had simillar issue, my phpserver was dropping some packages to dns and we had slow connection into db. The client you use can cache this. – Daniel Hornik Sep 23 '19 at 13:42
  • Does it work fast if you access the bucket using the [AWS Command-Line Interface (CLI)](http://aws.amazon.com/cli/) `aws s3 ls` command? – John Rotenstein Sep 23 '19 at 18:16
  • It works same slow with s3 cli tool. It gives error , something "interrupted with timeout", maybe s3 cli tool also has 30 seconds timeout. When i look with sniffer on s3cli traffic, i see that only small part of xml is received. Exactly same behavior like with my php curl app. The only tool i found working good is Cyberduck – Roman Gelembjuk Sep 24 '19 at 06:26
  • I tested the app from different location. The bucket is in default S3 region (east1) . I tested from US and from Europe locations. curl and php were same in all tests, but i think this doesn't matter as S3cli tool also has this problem – Roman Gelembjuk Sep 24 '19 at 06:28

1 Answers1

0

I have found the solution. If i add the argument "&versions=" to my request, all works fine! I don't understand this, but it is the fact

GET /?delimiter=%2F&marker=&max-keys=1000&prefix=&versions= HTTP/1.1

This works fine without any other changes.

UPDATE. This solution is not really a solution for me. I don't want to modify Amazon S3 API calls i do with that extra argument. I still hope to find some better solution, because this case looks like Amazon S3 has some broken server or equipment or configuration on some server

Roman Gelembjuk
  • 1,797
  • 2
  • 25
  • 50