CEPH pool df wrong metrics

professional_prutser · April 17, 2025, 7:28am

CMK version: Raw Edition 2.3.0p29
OS version: Debian Stable

deleted

r.sander · April 17, 2025, 7:41am

The check plugin included in Checkmk 2.3 adds “max_avail” and “bytes_used” for the total size and then uses “max_avail” as free space, from which the used percentage is then calculated.

github.com

Checkmk/checkmk/blob/release/2.3.0/cmk/base/plugins/agent_based/ceph_df_json_section.py#L34


      
                  "SUMMARY",
                  str_to_mebibyte(summary["total_bytes"]),
                  str_to_mebibyte(summary["total_avail_bytes"]),
                  0,
              )
          )
          mps.extend(
              [
                  (
                      pool["name"],
                      str_to_mebibyte(pool["stats"]["max_avail"])
                      + str_to_mebibyte(pool["stats"]["bytes_used"]),
                      str_to_mebibyte(pool["stats"]["max_avail"]),
                      0,
                  )
                  for pool in ceph_df["pools"]
              ]
          )
          return mps

The check plugin in Checkmk 2.4 uses “stored” instead of “bytes_used”, which is more correct:

github.com

Checkmk/checkmk/blob/v2.4.0b5/cmk/plugins/ceph/agent_based/cephdf.py#L89


      
          @dataclass(frozen=True)
          class Usage:
              used_mb: float
              avail_mb: float
              size_mb: float
          
              @classmethod
              def from_stats(cls, stats: Stats) -> Self:
                  if "stored" in stats:
                      # netto
                      used_mb = stats["stored"] / MIB
                      if (max_avail := stats.get("max_avail")) is not None and max_avail > 0:
                          return cls(
                              used_mb=used_mb,
                              avail_mb=(avail_mb := max_avail / MIB),
                              size_mb=avail_mb + used_mb,
                          )
          
                      if (perc_used := stats.get("percent_used")) is not None:
                          size_mb = used_mb / perc_used
                          return cls(

professional_prutser · April 17, 2025, 7:55am

deleted. I have no hope checkmk will do anything with feedback here

r.sander · April 17, 2025, 8:32am

Because the underlying function “df.df_check_filesystem_single” has “size_mb” and “avail_mb” as arguments and not percentage.
By using that function the team did not have to re-invent the wheel.

professional_prutser · April 17, 2025, 11:12am

As an end user i rather have ‘the team’ not to invent checks with bogus data. False positives are killing for the trust in an monitoring system. As an end user, i don’t care about underlying functions.

Without even a simple acknowledgment of this odd behavior, why would i even bother to give feedback to checkmk? Why would i ever bother testing a beta? Why would i ever bother to pay for this?