ceph-performance skillA
ceph-performance is agent-read markdown (skill) from air-gapped/skills: Tune and benchmark a healthy Ceph cluster, with Rook-on-Kubernetes specifics: mClock profiles and the NVMe capacity-fallback trap (osd_mclock_max_capacity_iops), osd_memory_target from pod requests, BlueStore allocation and cache defaults, PG autoscaler targets (mon_target_pg_per_osd, bulk, target_size_ratio), public/cluster network and Multus choices, msgr2 modes, Tentacle Fast EC (allow_ec_optimizations, stripe_unit), krbd vs rbd-nbd in ceph-csi, librbd cache, and benchmark method (rados bench.
Indexed from public GitHub and served as immutable, content-addressed versions. Install it pinned to an exact SHA-256 with the mdr CLI, and every file is verified against the hash recorded here before it reaches your agent. The deterministic audit below grades the latest version, and the same file always earns the same grade.
What the file says
# ceph-performance
Defaults read from `src/common/options/*.yaml.in` at v19.2.6 and v20.2.4,
docs.ceph.com (tentacle pages) and Rook docs at v1.20.7, on **2026-09-23**.
Measure before and after every change; `ceph config show osd.N <opt>` is
the value actually in force.
## mClock: the NVMe capacity trap
mClock sizes every reservation from each OSD's measured IOPS capacity. At
OSD start an `osd bench` sets `osd_mclock_max_capacity_iops_{hdd,ssd}`.
If the result exceeds the sanity threshold — **500 (HDD), 80 000 (SSD)** —
it is discarded and the OSD falls back to the default capacity
(**21 500** SSD) with a cluster-log warning. Modern NVMe routinely exceeds
80 000, so mClock under-drives it.
1. `ceph config show osd.N osd_mclock_max_capacity_iops_ssd` on every OSD;
`21500` means the fallback hit.
2. Measure the device with fio at 4 KiB random write, then
`ceph config set osd.N osd_mclock_max_capacity_iops_ssd <iops>`.
3. Re-check after replacing a device — the value is per OSD.
Profiles: `balanced` is the default from 17.2.7 and in Reef, Squid and
Tentacle (17.2.0–17.2.6 shipped `high_client_ops`). Use `high_client_ops`
…Read the whole file at its exact version.
How to install
mdr add air-gapped/skills/ceph-performance@git:20260922.6f1744dmdr add air-gapped/skills/ceph-performance@sha256:938c1d0f322c40c9Pin to a label to follow the author's releases, or to a sha256 to freeze the exact bytes forever. Either way the resolved hash is written to mdr.lock, and mdr install reproduces it on any machine.
[](https://markdownregistry.com/a/art_tjbucx6gfh65mnbz)
1 badge views in 30 days
Versions
| version | committed | commit | size | audit | |
|---|---|---|---|---|---|
| git:20260922.6f1744d latest | 2026-09-22 | 6f1744d | 6,943 B | A | view · diff |
| git:20260922.1fe1a44 | 2026-09-22 | 1fe1a44 | 7,274 B | A | view · diff |
| git:20260922.2273c5a | 2026-09-22 | 2273c5a | 7,192 B | A | view |
Audit of the latest version
- pass: Frontmatter block present
- pass: Frontmatter declares a name
- pass: Frontmatter declares a description
- pass: Size between 200 bytes and 200 KB (6943 bytes)
- pass: No zero-width or bidi control characters
- pass: No instruction hidden inside an HTML comment
- pass: No link to an exfiltration or paste host
- pass: No credential-shaped string
- pass: No instruction to send local credentials anywhere
- pass: No text hidden with inline styles
- pass: No prompt-injection phrasing
- pass: No curl or wget piped into a shell
- pass: No recursive delete of root, home or parent
- pass: No instruction to read or print local credentials
- pass: No base64 blob over 200 characters
- pass: No link to a raw IP address
- pass: No script tag
Source
air-gapped/skills · 5 stars · license MIT · pushed 2026-09-23 · branch main
API
GET https://markdownregistry.com/api/v1/artifacts/art_tjbucx6gfh65mnbz GET https://markdownregistry.com/api/v1/resolve?ref=air-gapped/skills/ceph-performance GET https://markdownregistry.com/api/v1/blob/938c1d0f322c40c972922378fc1bfaa25ab1bfa291b9467e6ce7540c21220061
Your agent does the legwork. You hear about the deals worth your word. Hand yours the standing instructions at modelranch.com and it joins the network that reads files like this one.