diff options
Diffstat (limited to 'src/web/api/queries/median')
| -rw-r--r-- | src/web/api/queries/median/README.md | 64 | ||||
| -rw-r--r-- | src/web/api/queries/median/median.c | 6 | ||||
| -rw-r--r-- | src/web/api/queries/median/median.h | 143 |
3 files changed, 213 insertions, 0 deletions
diff --git a/src/web/api/queries/median/README.md b/src/web/api/queries/median/README.md new file mode 100644 index 000000000..e6f6c04e7 --- /dev/null +++ b/src/web/api/queries/median/README.md @@ -0,0 +1,64 @@ +<!-- +title: "Median" +sidebar_label: "Median" +description: "Use median in API queries and health entities to find the 'middle' value from a sample, eliminating any unwanted spikes in the returned metrics." +custom_edit_url: https://github.com/netdata/netdata/edit/master/src/web/api/queries/median/README.md +learn_status: "Published" +learn_topic_type: "References" +learn_rel_path: "Developers/Web/Api/Queries" +--> + +# Median + +The median is the value separating the higher half from the lower half of a data sample +(a population or a probability distribution). For a data set, it may be thought of as the +"middle" value. + +`median` is not an accurate average. However, it eliminates all spikes, by sorting +all the values in a period, and selecting the value in the middle of the sorted array. + +Netdata also supports `trimmed-median`, which trims a percentage of the smaller and bigger values prior to finding the +median. The following `trimmed-median` functions are defined: + +- `trimmed-median1` +- `trimmed-median2` +- `trimmed-median3` +- `trimmed-median5` +- `trimmed-median10` +- `trimmed-median15` +- `trimmed-median20` +- `trimmed-median25` + +The function `trimmed-median` is an alias for `trimmed-median5`. + +## how to use + +Use it in alerts like this: + +``` + alarm: my_alert + on: my_chart +lookup: median -1m unaligned of my_dimension + warn: $this > 1000 +``` + +`median` does not change the units. For example, if the chart units is `requests/sec`, the result +will be again expressed in the same units. + +It can also be used in APIs and badges as `&group=median` in the URL. Additionally, a percentage may be given with +`&group_options=` to trim all small and big values before finding the median. + +## Examples + +Examining last 1 minute `successful` web server responses: + +-  +-  +-  +-  + +## References + +- <https://en.wikipedia.org/wiki/Median>. + + diff --git a/src/web/api/queries/median/median.c b/src/web/api/queries/median/median.c new file mode 100644 index 000000000..9865b485c --- /dev/null +++ b/src/web/api/queries/median/median.c @@ -0,0 +1,6 @@ +// SPDX-License-Identifier: GPL-3.0-or-later + +#include "median.h" + +// ---------------------------------------------------------------------------- +// median diff --git a/src/web/api/queries/median/median.h b/src/web/api/queries/median/median.h new file mode 100644 index 000000000..3d6d35925 --- /dev/null +++ b/src/web/api/queries/median/median.h @@ -0,0 +1,143 @@ +// SPDX-License-Identifier: GPL-3.0-or-later + +#ifndef NETDATA_API_QUERIES_MEDIAN_H +#define NETDATA_API_QUERIES_MEDIAN_H + +#include "../query.h" +#include "../rrdr.h" + +struct tg_median { + size_t series_size; + size_t next_pos; + NETDATA_DOUBLE percent; + + NETDATA_DOUBLE *series; +}; + +static inline void tg_median_create_internal(RRDR *r, const char *options, NETDATA_DOUBLE def) { + long entries = r->view.group; + if(entries < 10) entries = 10; + + struct tg_median *g = (struct tg_median *)onewayalloc_callocz(r->internal.owa, 1, sizeof(struct tg_median)); + g->series = onewayalloc_mallocz(r->internal.owa, entries * sizeof(NETDATA_DOUBLE)); + g->series_size = (size_t)entries; + + g->percent = def; + if(options && *options) { + g->percent = str2ndd(options, NULL); + if(!netdata_double_isnumber(g->percent)) g->percent = 0.0; + if(g->percent < 0.0) g->percent = 0.0; + if(g->percent > 50.0) g->percent = 50.0; + } + + g->percent = g->percent / 100.0; + r->time_grouping.data = g; +} + +static inline void tg_median_create(RRDR *r, const char *options) { + tg_median_create_internal(r, options, 0.0); +} +static inline void tg_median_create_trimmed_1(RRDR *r, const char *options) { + tg_median_create_internal(r, options, 1.0); +} +static inline void tg_median_create_trimmed_2(RRDR *r, const char *options) { + tg_median_create_internal(r, options, 2.0); +} +static inline void tg_median_create_trimmed_3(RRDR *r, const char *options) { + tg_median_create_internal(r, options, 3.0); +} +static inline void tg_median_create_trimmed_5(RRDR *r, const char *options) { + tg_median_create_internal(r, options, 5.0); +} +static inline void tg_median_create_trimmed_10(RRDR *r, const char *options) { + tg_median_create_internal(r, options, 10.0); +} +static inline void tg_median_create_trimmed_15(RRDR *r, const char *options) { + tg_median_create_internal(r, options, 15.0); +} +static inline void tg_median_create_trimmed_20(RRDR *r, const char *options) { + tg_median_create_internal(r, options, 20.0); +} +static inline void tg_median_create_trimmed_25(RRDR *r, const char *options) { + tg_median_create_internal(r, options, 25.0); +} + +// resets when switches dimensions +// so, clear everything to restart +static inline void tg_median_reset(RRDR *r) { + struct tg_median *g = (struct tg_median *)r->time_grouping.data; + g->next_pos = 0; +} + +static inline void tg_median_free(RRDR *r) { + struct tg_median *g = (struct tg_median *)r->time_grouping.data; + if(g) onewayalloc_freez(r->internal.owa, g->series); + + onewayalloc_freez(r->internal.owa, r->time_grouping.data); + r->time_grouping.data = NULL; +} + +static inline void tg_median_add(RRDR *r, NETDATA_DOUBLE value) { + struct tg_median *g = (struct tg_median *)r->time_grouping.data; + + if(unlikely(g->next_pos >= g->series_size)) { + g->series = onewayalloc_doublesize( r->internal.owa, g->series, g->series_size * sizeof(NETDATA_DOUBLE)); + g->series_size *= 2; + } + + g->series[g->next_pos++] = value; +} + +static inline NETDATA_DOUBLE tg_median_flush(RRDR *r, RRDR_VALUE_FLAGS *rrdr_value_options_ptr) { + struct tg_median *g = (struct tg_median *)r->time_grouping.data; + + size_t available_slots = g->next_pos; + NETDATA_DOUBLE value; + + if(unlikely(!available_slots)) { + value = 0.0; + *rrdr_value_options_ptr |= RRDR_VALUE_EMPTY; + } + else if(available_slots == 1) { + value = g->series[0]; + } + else { + sort_series(g->series, available_slots); + + size_t start_slot = 0; + size_t end_slot = available_slots - 1; + + if(g->percent > 0.0) { + NETDATA_DOUBLE min = g->series[0]; + NETDATA_DOUBLE max = g->series[available_slots - 1]; + NETDATA_DOUBLE delta = (max - min) * g->percent; + + NETDATA_DOUBLE wanted_min = min + delta; + NETDATA_DOUBLE wanted_max = max - delta; + + for (start_slot = 0; start_slot < available_slots; start_slot++) + if (g->series[start_slot] >= wanted_min) break; + + for (end_slot = available_slots - 1; end_slot > start_slot; end_slot--) + if (g->series[end_slot] <= wanted_max) break; + } + + if(start_slot == end_slot) + value = g->series[start_slot]; + else + value = median_on_sorted_series(&g->series[start_slot], end_slot - start_slot + 1); + } + + if(unlikely(!netdata_double_isnumber(value))) { + value = 0.0; + *rrdr_value_options_ptr |= RRDR_VALUE_EMPTY; + } + + //log_series_to_stderr(g->series, g->next_pos, value, "median"); + + g->next_pos = 0; + + return value; +} + +#endif //NETDATA_API_QUERIES_MEDIAN_H |