nathans wrote:
> [...] If you have time available, please have a read and send back
> any feedback you have [...]
It looks good. Some nits:
* page viii: as we discussed on IRC, "man -k performance" is too broad;
perhaps we should just link to "man PCPIntro".
* page 2, need motivation/mention for the archive folio concept
* page 7, pmdasystemd: This does not extract performance metrics, but
turns journal records into pcp events.
* page 7, mention pmdalogger: extracting lines of text from a log file
into pcp events
* page 7, pmwebd: Might mention HTTP beside JSON, just to spell out
the transport protocol.
* page 4...9: suggest moving list of tools after the 'conceptual
foundations' stuff
* page 11: "there may be at most one pmcd process": not technically true,
as people can run their own (on private ports, etc.); how about "normally"?
* page 14: "PMCS" - is there any need for this term? How about saying that
the pcp libraries provide utility functions for PMDAs? Or just use "PCP"
instead of "the PMCS" throughout the book?
* page 17: ${PCP_RC_DIR} appears unannounced. Refer to /etc/pcp.conf earlier?
* page 23: the "store" operation reminds me, we should take that permission
out of default pmcd.conf's.
* page 24, the "collector or monitor or both" question could be clearer.
I must admit I don't understand what exactly we're asking.
* page 27, "cannot connect to remote pmcd". Note firewalls.
* page 28, "netstat -a | grep 44321" presumes pmcd missing in /etc/services;
use "netstat -atn" instead.
* page 31, "dkvis" - whatchewtalkingaboutwillis?
* page 32, "The -h and -a options are mutually exclusive in all cases"
That is kind of sad actually; we should be willing to display a mixture
of historical and live data throughout.
* page 32. "/etc/pcp.*": those files should be listed far earlier, to explain
the use of ${PCP_FOOBAR} variables throughout the book. But actually,
is there some reason we don't autoconf-generate some fragment of this book,
so that actual per-installation path names are shown in the book? Maybe
everywhere, maybe in an appendix about This Very Installation?
* page 35, just a note, some neat libraries exist that take freeform english
to specify time points / intervals, like "last thursday, 3pm"; perhaps we
should pull them into libpcp for use with -S etc.
* page 36, just a note, PCP_COUNTER_WRAP seems like a hacky hammer; if
anything, it should be per-metric or something. If it's only used
in diagnostic emergencies, maybe not document it here?
* page 37, just a note, PMLOGGER_PORT, pmlc<->pmlogger really need some
authentication
* page 38, our mysterious superhero dkvis makes another appearance
* page 38, just a note, pmproxy and similar need outbound ACLs;
bug already filed
* page 39, the "-t interval" mention at the bottom looks weird word-wrapped
* page 49, the teaser use cases for pmie sound good; if those are actually
implementable today, we should provide a link to the pmie configuration
* page 54, top, backslash looks like \ not / :-)
* page 82, mysterious new caped superhero "pmview" makes an appearance
* page 92, add-on products shping / dbping being sweaty sidekicks to the heros;
as they're in the base package, why not drop "add-on products" vice "included
non-default PMDAs"?
* page 103, PMNS syntax, I wonder whether this part needs documenting. How
much do you expect a pcp user/administrator to read, never mind write such
a file?
* page 107, do we need to document the PDU concept?
(I above suggested dropping PMCS also.)
- FChE
|