Radoslaw Zarzynski

@athanatos: thanks for catching this! Just pushed a fixup.

> The two logic bugs are troubling for a PR you intend to quickly backport to squid. How did they make it through testing? As @markhpc points out, there was...

I'm looking into details of `ceph_test_rados`. I can see it issues sub-stripe reads **but always starting from the 0 chunk**. I implemented the `--offset` for `rados get` and poking with...

With the following patch that brings back the bug in `get_min_want_to_read_shards`: ```diff diff --git a/src/osd/ECCommon.cc b/src/osd/ECCommon.cc index e3aac98f3e2..03052e57204 100644 --- a/src/osd/ECCommon.cc +++ b/src/osd/ECCommon.cc @@ -320,16 +320,18 @@ void ECCommon::ReadPipeline::get_min_want_to_read_shards( {...

I hope this validates the testing methodology. Now I'm going to check whether similar problems can be triggered on earlier versions, without the recent fixups.

Manual testing ============= ``` Commit ID | Expected | Actual 46240812dfbde843ca6fad2381dc385553083317 | OK | OK 638e7060fd17131b3c95e8ae8da165c5a5163a14 | FAIL | FAIL 6f472a212a80a8e059647d16c31207bb8a3c4a83 | FAIL | FAIL 71be1923f49f830e8e62b1240fdd4c59179e2b5f | FAIL | FAIL...

I squashed the commits, manually tested for the bugs discovered during the review process and ensured we actually do partial reads (see below). @markhpc, @yuriw, @ljflores: I think we can...

I'm working on more test coverage for teuthology – pushed two early commits for `ceph_test_rados`. I'm afraid what we have in `main` is far from enough to test this PR....

The improved `ceph_test_rados` found another problem. Investigation showed it's actual defect in the EC partial reads, not in the modified tool itself, so I dissected the 2 commits into dedicated...

@ljflores, @yuriw: I pushed the fix for the problem unveiled by the modified testing tool. I think we're ready to build-and-test this PR (in parallel with the branch consisting purely...