Bug #64090
Updated by Anthony D'Atri 4 months ago
I have a cluster whose clients include Spark / Hadoop and Mimir. Initial deployment was with 17.2.5, all was well. Updated to 17.2.6 without a hitch. Updated to 17.2.7 and Spark / Hadoop broke with 400 errors. https://tracker.ceph.com/issues/17520 and https://tracker.ceph.com/issues/18965 are old tickets that would seem to be related. https://github.com/ceph/ceph/pull/53771 may be causal. We had to work around this by setting @spark.hadoop.fs.s3a.signing-algorithm=S3SignerType@ ` spark.hadoop.fs.s3a.signing-algorithm=S3SignerType`