Ok, am able to reproduce the issue now. The problem is that we send the target object for the operation to be garbage collected, and since we don't increase the refcount (being copied into itself) then the tail is actually removed. It takes a gc cycle for the object to lose its data. I pushed a couple of branches (wip-4776 and wip-4776-bobtail) with a fix. Thanks, Yehuda On Mon, Apr 22, 2013 at 9:21 AM, Yann ROBIN <yann.robin@xxxxxxxxxxxxx> wrote: > Hi, > > We use version 0.60. We use the .Net version of the AWS SDK. So basically what we do is to copyobject in same bucket with same key but we add metadata. > This only affects files greater than 512kb. > > var copyRequest = new CopyObjectRequest() > .WithDirective(S3MetadataDirective.REPLACE) > .WithSourceKey(key) > .WithSourceBucket(HtmlBucket) > .WithDestinationKey(key) > .WithDestinationBucket(HtmlBucket) > .WithMetaData(MetaHtmlLocations, metaLocation); > > -- > Yann > > -----Original Message----- > From: Yehuda Sadeh [mailto:yehuda@xxxxxxxxxxx] > Sent: lundi 22 avril 2013 16:45 > To: Yann ROBIN > Cc: ceph-users@xxxxxxxxxxxxxx > Subject: Re: Radosgw using s3 copy corrupt files > > On Mon, Apr 22, 2013 at 1:03 AM, Yann ROBIN <yann.robin@xxxxxxxxxxxxx> wrote: >> Hi, >> >> >> >> We use radosgw and s3 API and we recently needed to update metadata on >> some files. >> >> So we used the copy part of the S3 API for an in place replacement of >> the file adding some meta. >> >> >> >> We quickly saw some very high response time for some of those uploaded file. >> But there was no slow request. >> >> We looked at some file and saw that file greater than 512kb were corrupted. >> >> After 512kb the content of the files is : >> >> Status: 404 >> >> Content-Length: 75 >> >> Accept-Ranges: bytes >> >> Content-type: application/xml >> >> >> >> <?xml version="1.0" >> encoding="UTF-8"?><Error><Code>NoSuchKey</Code></Error> >> >> > > What version are you using? I tried to reproduce that but couldn't. I vaguely remember some fixes in that area that went in, but that was a while back. > > If you can reproduce it then rgw log with debug rgw = 20, debug ms = 1 could assist here. > > Thanks, > Yehuda > > _______________________________________________ ceph-users mailing list ceph-users@xxxxxxxxxxxxxx http://lists.ceph.com/listinfo.cgi/ceph-users-ceph.com
 |