Also correct the maximum length of the node name.
The callback name is "write_redis/%s" so the maximum
length of a node name is DATA_MAX_NAME_LEN - strlen("write_redis") -1.
src/utils_latency_test.c: Assure that large latency values also work.
The cdtime_t representation of 99s doesn't fit into 32bit anymore, thereby
assuring that cdtime_t is actually 64bit on platforms that currently
don't pass the test, i.e. EPEL {5,6} on i386.
CCLD test_utils_vl_lookup
Undefined first referenced
symbol in file
kstat_data_lookup daemon/.libs/libcommon.a(common.o)
kstat_lookup daemon/.libs/libcommon.a(common.o)
kstat_read daemon/.libs/libcommon.a(common.o)
getaddrinfo daemon/.libs/libcommon.a(common.o)
freeaddrinfo daemon/.libs/libcommon.a(common.o)
gai_strerror daemon/.libs/libcommon.a(common.o)
ld: fatal: symbol referencing errors. No output written to test_utils_vl_lookup
collect2: error: ld returned 1 exit status
This hopefully fixes "make check" on Solaris, which currently fails
with:
CC common_test.o
CCLD test_common
Undefined first referenced
symbol in file
kc ./.libs/libcommon.a(common.o)
ld: fatal: symbol referencing errors. No output written to test_common
collect2: error: ld returned 1 exit status
src/utils_latency.c: Store "bin_width" as cdtime_t.
This solves the integer overflow when passing huge latency values to
latency_counter_add(). In addition to fixing the overflow the function
will now ignore values that are larger than LLONG_MAX, i.e. the longest
possible latency is 272 years. As a nice side-effect, the precission of
latency_counter_get_percentile() is improved.
Testing: Return failure from OK1() and other fixes.
* The tests for common and meta_data logged "not ok" but didn't signal
failure because OK1() didn't include a return(-1) line. Adding this line
caused some restructuring of the utils_vl_lookup test, because it used
that macro in non-int functions.
* Fix DBLEQ() to work correctly with an expected NaN. Previously, the
if condition would fall through to the "expect != actual" part, which
is true for "NaN != NaN".
* Let the mock cdtime() return a non-zero value, as the (invalid) zero
value is used in parse_values() to detect whether the time has been
parsed already. This lead to the "N:..." tests failing.
* Correct the expected behavior of meta_data_add_*() when keys already
exist: they're replaced rather than causing an error.
daemon: Clean up the Upstart and Systemd notification code.
Document the use of "Linux abstract namespace sockets" and use sendto(2)
instead of sendmsg(2) to avoid having to initialize struct msghdr
and struct iovec. Removes the non-portable bzero() calls.
Ruud van Melick [Tue, 30 Jun 2015 16:08:49 +0000 (16:08 +0000)]
Fix Perl errors
"my" variable $cmd masks earlier declaration in same scope at /tmp/contrib_collectd2html.pl line 223.
syntax error at /tmp/contrib_collectd2html.pl line 219, near "$span qw(1hour 1day 1week 1month)"
Global symbol "$span" requires explicit package name at /tmp/contrib_collectd2html.pl line 221.
Global symbol "$span" requires explicit package name at /tmp/contrib_collectd2html.pl line 224.
Global symbol "$span" requires explicit package name at /tmp/contrib_collectd2html.pl line 225.
Global symbol "$span" requires explicit package name at /tmp/contrib_collectd2html.pl line 232.
Global symbol "$span" requires explicit package name at /tmp/contrib_collectd2html.pl line 237.
syntax error at /tmp/contrib_collectd2html.pl line 245, near "}"
Florian Forster [Thu, 20 Nov 2014 17:25:21 +0000 (18:25 +0100)]
mqtt plugin: Change message format to one topic per metric.
The payload now includes all data sources and the timestamp. The payload
is formatted with the existing format_values() function, which removed
quite a couple of lines of code.
Florian Forster [Thu, 20 Nov 2014 16:42:42 +0000 (17:42 +0100)]
mqtt plugin: Compile fixes.
This compiles with libmosquitto 0.15.0. It appears newer versions of the
library have some interface changes, so we should add a version check.
Luckily, libmosquitto provides everything we need.
The previous implementation broke when off1 / off2 were outside of "string"
but within "buflen". It also had problems if the replacement string was
too long. This new implementation truncates the buffer as expected and
is properly tested.
Florian Forster [Mon, 29 Jun 2015 10:50:51 +0000 (12:50 +0200)]
src/daemon/common.[ch]: Improve signature of value_to_rate().
Passing in "value" as a derive_t and returning the rate as a value_t is
just inconvenient and wrong. Instead, pass "value" as a value_t and return
the rate in a gauge_t. While at it, move the state pointer to the end of
the argument list.