bench event log performance

[jaychien1103]

I use the phosphor snmp trap to send event, and also adding event into logging service.
The event in my platform ast2520 and flash is SPI flash with 50MHz.
I stress the logging, it just can log 8 events in a second.
and when the event achieve to 3000 items, stop the event creation,
webbrowser need about 25 second to show the event and when the target running top during the web connect event log
page, sometimes it cannot display over 3 minutes.
the top can see the bmcweb and phosphor-logging have higher CPU usage.
and also each event occupied a file.
Does anyone have benchmark of this module ?

thanks

[Palaniappanarunachalam]

Most of this is fixed now - lg2::commit removed the journalSync+ journal scan that made log creation slow on the old path, and error_cap limits how many entries are kept.

One small leftover: each entry is still written to flash twice (cereal + JSON). Going JSON-only would roughly halve that, but it needs RBMC sync to move to JSON first, so noting it for later rather than now.

[williamspatrick]

One small leftover: each entry is still written to flash twice (cereal + JSON). Going JSON-only would roughly halve that, but it needs RBMC sync to move to JSON first, so noting it for later rather than now.

We should add a meson option to disable cereal. That can be used for platforms that do not need backwards compatibility.

We might want to frame it as a "stop writing cereal" but leave the code to ingest it (and maybe delete it after consumed). This gives a clear upgrade path for phosphor-logging.