Python 3 compatibility fixes and two new features

[xorpaul]

Summary

• Python 3 compatibility — fix #!/usr/bin/env python shebang and stderr bytes-decoding bugs across all ten scripts; without these fixes the plugins either crash with TypeError or silently print b'...' byte-string representations in Nagios output on Python 3.12+
• check_ceph_osd rewrite — replace the broken IP-based ceph osd dump approach with ceph osd tree --format json; the old code failed on clusters with separate storage networks where the DNS name resolves to a management IP that never appears in the OSD dump
• check_ceph_rgw: add --keyring — radosgw-admin accepts --keyring like all other Ceph tools but the plugin had no way to pass it
• check_ceph_mds: add --min-active — cephadm-managed clusters assign ephemeral random suffixes to MDS daemon names, making the existing --name mode unusable for stable Nagios checks; the new mode checks that a filesystem has at least N active daemons instead

Detailed changes

Python 3 compatibility (all scripts)

All scripts updated to #!/usr/bin/env python3.

subprocess.communicate() returns stderr as bytes in Python 3. Three scripts called .split('\n') directly on the bytes object, crashing with TypeError whenever Ceph writes to stderr:

• src/check_ceph_df
• src/check_ceph_health
• src/check_ceph_osd_df

Three more scripts printed err via %s / .format() without decoding, causing error messages to include the raw b'' wrapper:

• src/check_ceph_mgr
• src/check_ceph_mon
• src/check_ceph_rgw (also fixed in the --keyring commit)

All fixed with .decode('utf-8', errors='replace') before any string operation.

check_ceph_osd — switch from ceph osd dump to ceph osd tree --format json

The previous implementation resolved the host argument via DNS, then regex-searched the resulting IP in ceph osd dump text output. This fails on any cluster where the management/DNS IP differs from the storage-network IP the OSD daemons actually bind to — a common production setup.

The new implementation uses ceph osd tree --format json, which encodes the CRUSH host-to-OSD mapping directly. The host lookup tries the FQDN first, then the short hostname, so CRUSH bucket names (typically short, e.g. osd2) are matched correctly even when -H receives an FQDN like osd2.example.com.

Additional fixes in the same script:

• Remove socket / DNS resolution code — no longer needed
• Remove invalid escape sequences '\.', '\[', '\]' in src/check_ceph_osd that triggered SyntaxWarning in Python 3.12+
• --osdid now matches the numeric OSD ID (e.g. --osdid 3) rather than requiring the osd.3 prefix

check_ceph_rgw — add --keyring

Add -k/--keyring with the same existence-check validation and command-line wiring used by all other plugins in the collection. Without this, users on hosts without a default keyring had no way to authenticate.

check_ceph_mds — add --min-active

With cephadm, MDS daemon names include ephemeral random IDs (format: <fs>.<host>.<random>, e.g. m01.mds2.fezsws). These change on every daemon restart, making the --name mode unsuitable for permanent Nagios service definitions.

The new -A/--min-active <N> mode checks that the given filesystem has at least N daemons in up:active state without a laggy flag. Behaviour:

• Active count ≥ N, no laggy daemons → OK
• Active count ≥ N, but some laggy → WARNING
• Active count < N → CRITICAL

--name is unchanged and still works for environments with stable MDS names. Exactly one of --name or --min-active must be provided.

Testing

Verified against a Pacific/Quincy cluster with 12 CephFS filesystems, 96 OSDs across 4 hosts, and separate management / storage networks:

check_ceph_health  --id nagios --keyring /var/lib/nagios/ceph.client.nagios.keyring
→ HEALTH OK

check_ceph_osd     --id nagios --keyring /var/lib/nagios/ceph.client.nagios.keyring -H osd2.example.com
→ OSD OK / Up OSDs: osd.3 osd.8 … (24 OSDs)

check_ceph_mds     --id nagios --keyring /var/lib/nagios/ceph.client.nagios.keyring --filesystem m01 --min-active 5
→ MDS OK: FS 'm01' has 5 active MDS