B.20 loss.*
Every field the catalogue defines under loss, in path order. Each entry gives the field's definition and, under Carried by, every event type that writes it.
B.20.1 loss.bytes #
- Type:
uint.bytes - Asserted: no
- Carried in: the payload
- Defined in:
kmes.evman
The bytes of events a loss discarded. Reported in bytes where KMES cannot count events: when a ring's framing is found corrupt, everything pending is dropped at once without being parsed, and when a capacity swap skips an event too large for the smaller ring, only its size is known.
Carried by:
No event carries this field yet.
B.20.2 loss.class #
- Type:
uint.enum - Values:
0 userspace·1 kmes·2 kacs·3 lcs·4 ntfe - Set: open
- Asserted: no
- Carried in: the payload
- Defined in:
kmes.evman
The emission path of the events that were lost, numbered as emitter.class
numbers it: the kernel subsystem, or userspace, that was writing them. A
loss record is written by KMES, so its own emitter.class is always
kmes; this is the field that says whose events went missing.
Carried by:
No event carries this field yet.
B.20.3 loss.count #
- Type:
uint - Asserted: no
- Carried in: the payload
- Defined in:
kmes.evman
The number of events lost in one occurrence. For a gap found in the
sequence numbers, the size of that gap: loss.sequence-last minus
loss.sequence, plus one.
Carried by:
B.20.4 loss.preceding-time #
- Type:
uint.time - Asserted: no
- Carried in: the payload
- Defined in:
kmes.evman
The time of the last record received before a gap, in nanoseconds since the Unix epoch. With the time of the record that revealed the gap, it bounds when the lost events were written. Absent when no record preceded the gap on that ring, as at the start of a boot.
Carried by:
B.20.5 loss.sequence #
- Type:
uint - Asserted: no
- Carried in: the payload
- Defined in:
kmes.evman
The first sequence number a loss covers. Sequence numbers are contiguous
per CPU ring, so with buffer.cpu this locates the gap exactly, and it is
the field that reconciles a loss KMES reports with a gap a consumer infers
from event.sequence.
Not every loss has one. An event dropped because its CPU had no usable ring, an event emitted before KMES started, and an emit refused at the system call consume no sequence number and leave no gap to point at.
Carried by:
B.20.6 loss.sequence-last #
- Type:
uint - Asserted: no
- Carried in: the payload
- Defined in:
kmes.evman
The last sequence number of a gap, where loss.sequence is the first.
Both ends are included in the loss.
Carried by:
B.20.7 loss.total #
- Type:
uint - Asserted: no
- Carried in: the payload
- Defined in:
kmes.evman
The cumulative count of events a ring has dropped. Not emitted today. KMES keeps the counter, but it is not in the ring's metadata page and nothing outside the kernel's own tests reads it, so no emitter has it to report. It also under-reports: a corrupt-ring discard, an event skipped by a capacity swap and an event lost for want of a usable ring are never added to it, so even once it is emitted it is a lower bound rather than a total.
Carried by:
No event carries this field yet.
Generated from kmes.evman by pkm/tools/gen-events-book.py. Edit the fragment, not this page.