lveinfo and lvestats2

[brooks]

Understanding the granularity of uCPU

[brooks]

The new data available fr om lvestats2 is excellent. Thank you.

The data allows me to better analyze the over-all resource utilization for the server and for a specific site. For example, to find any faults over the last 24 hours I might run:

# lveinfo --period 24h --by-fault any --show-columns id from to iopsf iof cpuf epf pmemf mcpu ucpu uep upmem

(this would make a nice format for the summary email, the current format is too long)

+----+-----------+-----------+-----+---+----+---+-----+----+----+----+-----+

| ID |    From   |     To    |IOPSf|IOf|CPUf|EPf|PMemF|mCPU|uCPU|uEP |uPMem|

+----+-----------+-----------+-----+---+----+---+-----+----+----+----+-----+

|1094|12-18 16:15|12-19 16:15|0    |1  |0   |0  |0    |30  |0.0%|0.0%|0.0% |

|1095|12-18 16:15|12-19 16:15|0    |0  |1   |0  |0    |100 |0.0%|0.0%|0.1% |

+----+-----------+-----------+-----+---+----+---+-----+----+----+----+-----+

To drill down into the CPU usage for LVE 1095 I might use:

# lveinfo --period 24h --id 1095 --show-columns id from to cpuf mcpu ucpu | more

|1095|12-19 13:18|12-19 13:19|0   |87  |87.0%|

|1095|12-19 13:19|12-19 13:20|1   |100 |20.4%|

|1095|12-19 13:21|12-19 13:22|0   |8   |8.4% |

Youll notice the uCPU field is only 20.4% when the fault occurred. Unless lvestats2 is doing sub-minute granularity shouldnt the value of uCPU be 100% when the LVE hit the CPU lim it? Other than during this fault period the uCPU seems to match the value of mCPU (as a percentage).

[brooks]

Here I see the opposite. I display the entry proc stats for LVE1004 for the last 5 minutes:

# lveinfo --period 5m --id 1004 --show-columns id from to epf mep uep

+----+-----------+-----------+---+---+----+

| ID |    From   |     To    |EPf|mEP|uEP |

+----+-----------+-----------+---+---+----+

|1004|12-19 17:58|12-19 17:59|0  |0  |0.3%|

|1004|12-19 17:59|12-19 18:00|0  |0  |0.4%|

|1004|12-19 18:00|12-19 18:01|0  |0  |0.0%|

|1004|12-19 18:01|12-19 18:02|0  |0  |0.0%|

|1004|12-19 18:02|12-19 18:03|0  |0  |0.3%|

+----+-----------+-----------+---+---+----+

mEP = 0 when there has been 1 or 2 lsphp process running as that user over that same period of time. It looks like the uEP stats reflect that yet mEP doesnt.