Apache Ozone

Pinned 2.2.1 gateway and arm64 block/restart/CAS/range evidence exist; the latest exact-tip hosted packet is diagnostic after hard 1,000-IOPS misses, while customer topology, backup/DR, secure tenancy, and release gates remain external. The current maturity is Experimental: the real arm64 gateway harness has the block, CAS, range, failure, restart/reopen, and cleanup checkpoint above. Its SQLite/PGlite compositions passed through the real Ozone gateway, and a separate single-node TiDB/Ozone run passed the direct TiDB contract, ChunkedFs partial/truncate/CAS/stale-fencing/ reopen, ambiguous-commit, and cleanup checks. The arm64 Node provider and Node/Rust CLI run recorded pass=7 skip=1 fail=0. A durable FoundationDB/Ozone run also passed with three fixed-address FoundationDB 7.4.7 servers, persistent volumes, double/SSD configuration, transaction readiness across a replicated-node restart, fresh-client reopen, and owned cleanup. A dedicatedozone-tidb hosted job is wired for durable v8.5.7 TiDB with Node 24 against the real Ozone gateway, but the current replacement job is queued and has no result yet. A dedicated hostedozone-compositions job now installs PGlite, builds the public Node addon, and runs the real SQLite/PGlite mixed-metadata gate. Its opt-in consumer phase covers the Node provider matrix, Node CLI, and Rust CLI against live Ozone/PGlite/R2-compatible services, but the hosted result is still pending and revision-specific. Durable multi-node TiDB, secure/replicated Ozone deployment, and the broader backend matrix are not yet accepted. The retained W26 packet on9c098e5 / hosted run 35585066458 remains the last accepted scoped result. The newer hosted packet35635486040 completed all four configured provider lifecycles but failed the 1,000-IOPS gate; aggregate job106458415293 rejected the missing pass marker, so its artifacts are diagnostic only. W26's credential-free rollout contract fixes the Tier 1 99.99% availability objective, 1,000 IOPS per drive, five-minute RPO/RTO, TLS/SigV4, tenant-scoped prefixes, and three-node/three-replica/three-domain topology; these are customer/Ozone requirements, not proof of a deployed environment. Replacement run 35641941218 is terminal for W26 even though unrelated native jobs kept the parent workflow active. Its producer jobs 106473112920, 106473112915, and106473112409, plus aggregate106477164582, failed closed on the hard IOPS gate. The local overlap/read/shutdown regression suite is green, but this first remediation is insufficient; secure customer topology, backup/DR, measured SLO/capacity, and release/canary/rollback remain open. The current publication-barrier implementation adds a conservative provider capability that skips only the redundant post-publish flush probe while retaining explicit syncfs; the focused ChunkedFs suite passes 20/20 and the full locked workspace/Clippy checks are green. Fresh hosted run 35672928117 targets exact SHA fbc8346d, but its Ozone producers were queued or in progress and no aggregate existed at the recorded snapshot, so it is not acceptance evidence. The later terminal diagnostic packet35674425514 completed 1,200/1,200 lifecycle operations with zero timeouts and cleanup failures for every provider row, but measured only 637.01 IOPS for SQLite/R2, 457.65 for PGlite/R2, 147.31 for TiDB/R2, and 126.50 for FoundationDB/R2 against the hard 1,000-IOPS target; its aggregate failed closed without pass markers. The newer manual run 35683158821 targeted exact SHA14dbf2c6 and is now terminal but diagnostic: SQLite/R2 measured 754.59 IOPS, PGlite/R2 817.10, TiDB/R2 143.04, and FoundationDB/R2 345.17against the unchanged 1,000-IOPS target. Each provider completed 1,200/1,200 lifecycle operations with zero timeouts and cleanup failures, but the provider jobs failed the hard threshold and the aggregate job 106606577835 failed closed on missingOZONE_IOPS_PASS markers. No acceptance or production claim is made from this run. The newer terminal packet 35686340751 selected exact SHA1e64bc25: base Ozone passed, SQLite/R2 measured997.06 IOPS, PGlite/R2 592.43, and TiDB/R2277.59; FoundationDB produced no benchmark after its locked-preflight failure, and aggregate job106616062969 failed closed. A fresh manual dispatch35691451007 selected exact SHA dccd8351 and remains diagnostic: SQLite/R2 measured 213.947686 IOPS, PGlite/R2 2065.669446, TiDB/R2 333.356025, and FoundationDB/R2 363.254472. Every row completed 1,200/1,200 lifecycle operations with zero timeouts and cleanup failures. PGlite cleared the row target, but SQLite, TiDB, and FoundationDB missed the hard 1,000 IOPS threshold, so the aggregate failed closed without OZONE_IOPS_PASS. No acceptance or production claim is made from this run; the next work is correctness-preserving variance diagnosis and requalification. The latest terminal packet 35698854392 at exact source116e9ed4 passed the Ozone base checks and recorded SQLite/R2 at 1051.976655 IOPS, but PGlite/R2 measured837.779225, TiDB/R2 463.080116, and FoundationDB/R2 450.696578 against the unchanged hard1000-IOPS target. Each row completed 1,200/1,200 operations with zero timeouts and cleanup failures; the aggregate failed closed without OZONE_IOPS_PASS. This is diagnostic provider evidence, not an Ozone acceptance or production claim.