Hi John, Thanks for the info. I'm very curious to see what's in the haproxy logs with regard to TCP. -- Luke Bakken Engineer lbak...@basho.com
On Wed, Jan 13, 2016 at 3:50 PM, John Fanjoy <jfan...@inetu.net> wrote: > Luke, > > As a test I’ve already increased all timeouts to 5 minutes but the failure > occurs within under 1 minute so it doesn’t appear to be timeout related. I > change the logs to tcplog tomorrow and let you know if I find anything. > > Thanks > > John Fanjoy > Systems Engineer > jfan...@inetu.net > > > > > > On 1/13/16, 6:05 PM, "Luke Bakken" <lbak...@basho.com> wrote: > >>haproxy ships with some "short" default timeouts. If CyberDuck is able >>to upload these files faster than aws-sdk, it may be doing so within >>the default haproxy timeouts. >> >>You can also look at haproxy's log to see if you find any TCP >>connections that it has closed. >>-- >>Luke Bakken >>Engineer >>lbak...@basho.com >> >> >>On Wed, Jan 13, 2016 at 3:02 PM, John Fanjoy <jfan...@inetu.net> wrote: >>> Luke, >>> >>> I may be able to do that. The only problem is without haproxy I have no way >>> to inject CORS headers which the browser requires, but I may be able to >>> write up small nodejs app to get past that and see if it is somehow related >>> to haproxy. The fact that these errors are not present when using Cyberduck >>> which is also talking to haproxy leads me to believe that’s not the cause, >>> but it’s definitely worth testing. >>> >>> -- >>> John Fanjoy >>> Systems Engineer >>> jfan...@inetu.net >>> >>> >>> >>> >>> >>> On 1/13/16, 5:55 PM, "Luke Bakken" <lbak...@basho.com> wrote: >>> >>>>John - >>>> >>>>The following error indicates that the connection was unexpectedly >>>>closed by something outside of Riak while the chunk is uploading: >>>> >>>>{badmatch,{error,closed}} >>>> >>>>Is it possible to remove haproxy to test using the the aws-sdk? >>>> >>>>That is my first thought as to the cause of this issue, especially >>>>since writing to S3 works with the same code. >>>> >>>>-- >>>>Luke Bakken >>>>Engineer >>>>lbak...@basho.com >>>> >>>>On Wed, Jan 13, 2016 at 2:46 PM, John Fanjoy <jfan...@inetu.net> wrote: >>>>> Luke, >>>>> >>>>> Yes on both parts. To confirm cyberduck was using multi-part I actually >>>>> tailed the console.log while it was uploading the file, and it uploaded >>>>> the file in approx. 40 parts. Afterwards the parts were reassembled as >>>>> you would expect. The AWS-SDK for javascript has an object called >>>>> ManagedUpload which automatically switches to multi-part when the input >>>>> is larger than the maxpartsize (default 5mb). I have confirmed that it is >>>>> splitting the files up, but so far I’ve only ever seen one part get >>>>> successfully uploaded before the others failed at which point it removes >>>>> the upload (DELETE call) automatically. I also verified that the >>>>> javascript I have in place does work with an actual AWS S3 bucket to rule >>>>> out coding issues on my end and the same >400mb file was successfully >>>>> uploaded to the bucket I created there without issue. >>>>> >>>>> A few things worth mentioning that I missed before. I am running riak-s2 >>>>> behind haproxy. Haproxy is handling ssl and enabling CORS for browser >>>>> based requests. I have tested smaller files (~4-5mb) and GET requests >>>>> using the browser client and everything works with my current haproxy >>>>> configuration, but the larger files are failing, usually after 1 part is >>>>> successfully uploaded. I can also list bucket contents and delete >>>>> existing contents. The only feature that is not working appears to be the >>>>> multi-part uploads. We are running centOS 7 (kernel version >>>>> 3.10.0-327.4.4.el7.x86_64). Please let me know if you have any further >>>>> questions. >>>>> >>>>> -- >>>>> John Fanjoy >>>>> Systems Engineer >>>>> jfan...@inetu.net _______________________________________________ riak-users mailing list riak-users@lists.basho.com http://lists.basho.com/mailman/listinfo/riak-users_lists.basho.com