Hi Ceph Community, I am writing to inquire about a specific behavior change regarding data consistency in Erasure Coded pools between Ceph Tentacle v20.1.0 and v20.2.0. I previously encountered a deterministic data corruption issue during recovery in v20.1.0. After upgrading to v20.2.0 and repeating the same test, the issue appears to be resolved. I am interested in understanding the technical logic behind this fix. Here are the details of my test scenario: Cluster Configuration Version: Tested on Tentacle v20.1.0 (Issue observed) vs. v20.2.0 (Issue resolved) Layout: 8 OSD hosts EC Profile: k=6, m=2 (8 total shards) min_size: Manually lowered from 7 to 6 (k) Test Scenario (Reproduction Steps) 1. Client Setup: Created RBD images, formatted with ext4, and mounted them on 3 separate clients (1 TB image per client). 2. I/O Start: Started writing data patterns to the mount points until the mounted filesystems were filled to capacity. 3. Failure Injection: While writing, I forcibly brought down 2 OSD hosts. 4. Write Completion: Since min_size=6, writes continued and succeeded in a degraded state (Acting set size = 6). 5. Pre-Recovery Verification: Before bringing the nodes back up, I verified the data (MD5 checksum). - Result: Data was intact/readable (likely reading the 6 available systematic chunks). 6. Recovery: I brought the 2 down OSD hosts back up. The cluster performed recovery/peering and reached Active+Clean. 7. Post-Recovery Verification: After the cluster became Clean, I performed the integrity check again   The Observation On v20.1.0 MD5 Mismatch. The data became corrupted after recovery. This happened with 100/100 files tested. On v20.2.0 No data loss or corruption. MD5 checksums matched perfectly after recovery. It is great to see that v20.2.0 has resolved this critical issue. I am curious to learn what caused the corruption in the earlier version and how the recovery logic has been improved to handle this scenario correctly. Could anyone share insights into the root cause or the nature of the fix? If there are specific Pull Requests or Issue tracker links available for reference, I would greatly appreciate it. Thank you for your time and assistance. Best regards, Chanu Yu