0
0
Fork 0
mirror of https://github.com/go-gitea/gitea synced 2024-11-16 15:01:07 +01:00
gitea/vendor/github.com/olivere/elastic/v7/cat_shards.go
6543 12a1f914f4
Vendor Update Go Libs (#13166)
* update github.com/alecthomas/chroma v0.8.0 -> v0.8.1

* github.com/blevesearch/bleve v1.0.10 -> v1.0.12

* editorconfig-core-go v2.1.1 -> v2.3.7

* github.com/gliderlabs/ssh v0.2.2 -> v0.3.1

* migrate editorconfig.ParseBytes to Parse

* github.com/shurcooL/vfsgen to 0d455de96546

* github.com/go-git/go-git/v5 v5.1.0 -> v5.2.0

* github.com/google/uuid v1.1.1 -> v1.1.2

* github.com/huandu/xstrings v1.3.0 -> v1.3.2

* github.com/klauspost/compress v1.10.11 -> v1.11.1

* github.com/markbates/goth v1.61.2 -> v1.65.0

* github.com/mattn/go-sqlite3 v1.14.0 -> v1.14.4

* github.com/mholt/archiver v3.3.0 -> v3.3.2

* github.com/microcosm-cc/bluemonday 4f7140c49acb -> v1.0.4

* github.com/minio/minio-go v7.0.4 -> v7.0.5

* github.com/olivere/elastic v7.0.9 -> v7.0.20

* github.com/urfave/cli v1.20.0 -> v1.22.4

* github.com/prometheus/client_golang v1.1.0 -> v1.8.0

* github.com/xanzy/go-gitlab v0.37.0 -> v0.38.1

* mvdan.cc/xurls v2.1.0 -> v2.2.0

Co-authored-by: Lauris BH <lauris@nix.lv>
2020-10-16 01:06:27 -04:00

389 lines
17 KiB
Go
Vendored

// Copyright 2012-present Oliver Eilhard. All rights reserved.
// Use of this source code is governed by a MIT-license.
// See http://olivere.mit-license.org/license.txt for details.
package elastic
import (
"context"
"fmt"
"net/http"
"net/url"
"strings"
"github.com/olivere/elastic/v7/uritemplates"
)
// CatShardsService returns the list of shards plus some additional
// information about them.
//
// See https://www.elastic.co/guide/en/elasticsearch/reference/7.6/cat-shards.html
// for details.
type CatShardsService struct {
client *Client
pretty *bool // pretty format the returned JSON response
human *bool // return human readable values for statistics
errorTrace *bool // include the stack trace of returned errors
filterPath []string // list of filters used to reduce the response
index []string
bytes string // b, k, kb, m, mb, g, gb, t, tb, p, or pb
local *bool
masterTimeout string
columns []string
time string // d, h, m, s, ms, micros, or nanos
sort []string // list of columns for sort order
headers http.Header
}
// NewCatShardsService creates a new CatShardsService.
func NewCatShardsService(client *Client) *CatShardsService {
return &CatShardsService{
client: client,
}
}
// Pretty tells Elasticsearch whether to return a formatted JSON response.
func (s *CatShardsService) Pretty(pretty bool) *CatShardsService {
s.pretty = &pretty
return s
}
// Human specifies whether human readable values should be returned in
// the JSON response, e.g. "7.5mb".
func (s *CatShardsService) Human(human bool) *CatShardsService {
s.human = &human
return s
}
// ErrorTrace specifies whether to include the stack trace of returned errors.
func (s *CatShardsService) ErrorTrace(errorTrace bool) *CatShardsService {
s.errorTrace = &errorTrace
return s
}
// FilterPath specifies a list of filters used to reduce the response.
func (s *CatShardsService) FilterPath(filterPath ...string) *CatShardsService {
s.filterPath = filterPath
return s
}
// Header adds a header to the request.
func (s *CatShardsService) Header(name string, value string) *CatShardsService {
if s.headers == nil {
s.headers = http.Header{}
}
s.headers.Add(name, value)
return s
}
// Headers specifies the headers of the request.
func (s *CatShardsService) Headers(headers http.Header) *CatShardsService {
s.headers = headers
return s
}
// Index is the name of the index to list (by default all indices are returned).
func (s *CatShardsService) Index(index ...string) *CatShardsService {
s.index = index
return s
}
// Bytes represents the unit in which to display byte values.
// Valid values are: "b", "k", "kb", "m", "mb", "g", "gb", "t", "tb", "p" or "pb".
func (s *CatShardsService) Bytes(bytes string) *CatShardsService {
s.bytes = bytes
return s
}
// Local indicates to return local information, i.e. do not retrieve
// the state from master node (default: false).
func (s *CatShardsService) Local(local bool) *CatShardsService {
s.local = &local
return s
}
// MasterTimeout is the explicit operation timeout for connection to master node.
func (s *CatShardsService) MasterTimeout(masterTimeout string) *CatShardsService {
s.masterTimeout = masterTimeout
return s
}
// Columns to return in the response.
//
// To get a list of all possible columns to return, run the following command
// in your terminal:
//
// Example:
// curl 'http://localhost:9200/_cat/shards?help'
//
// You can use Columns("*") to return all possible columns. That might take
// a little longer than the default set of columns.
func (s *CatShardsService) Columns(columns ...string) *CatShardsService {
s.columns = columns
return s
}
// Sort is a list of fields to sort by.
func (s *CatShardsService) Sort(fields ...string) *CatShardsService {
s.sort = fields
return s
}
// Time specifies the way that time values are formatted with.
func (s *CatShardsService) Time(time string) *CatShardsService {
s.time = time
return s
}
// buildURL builds the URL for the operation.
func (s *CatShardsService) buildURL() (string, url.Values, error) {
// Build URL
var (
path string
err error
)
if len(s.index) > 0 {
path, err = uritemplates.Expand("/_cat/shards/{index}", map[string]string{
"index": strings.Join(s.index, ","),
})
} else {
path = "/_cat/shards"
}
if err != nil {
return "", url.Values{}, err
}
// Add query string parameters
params := url.Values{
"format": []string{"json"}, // always returns as JSON
}
if v := s.pretty; v != nil {
params.Set("pretty", fmt.Sprint(*v))
}
if v := s.human; v != nil {
params.Set("human", fmt.Sprint(*v))
}
if v := s.errorTrace; v != nil {
params.Set("error_trace", fmt.Sprint(*v))
}
if len(s.filterPath) > 0 {
params.Set("filter_path", strings.Join(s.filterPath, ","))
}
if s.bytes != "" {
params.Set("bytes", s.bytes)
}
if s.time != "" {
params.Set("time", s.time)
}
if v := s.local; v != nil {
params.Set("local", fmt.Sprint(*v))
}
if s.masterTimeout != "" {
params.Set("master_timeout", s.masterTimeout)
}
if len(s.columns) > 0 {
// loop through all columns and apply alias if needed
for i, column := range s.columns {
if fullValueRaw, isAliased := catShardsResponseRowAliasesMap[column]; isAliased {
// alias can be translated to multiple fields,
// so if translated value contains a comma, than replace the first value
// and append the others
if strings.Contains(fullValueRaw, ",") {
fullValues := strings.Split(fullValueRaw, ",")
s.columns[i] = fullValues[0]
s.columns = append(s.columns, fullValues[1:]...)
} else {
s.columns[i] = fullValueRaw
}
}
}
params.Set("h", strings.Join(s.columns, ","))
}
if len(s.sort) > 0 {
params.Set("s", strings.Join(s.sort, ","))
}
return path, params, nil
}
// Do executes the operation.
func (s *CatShardsService) Do(ctx context.Context) (CatShardsResponse, error) {
// Get URL for request
path, params, err := s.buildURL()
if err != nil {
return nil, err
}
// Get HTTP response
res, err := s.client.PerformRequest(ctx, PerformRequestOptions{
Method: "GET",
Path: path,
Params: params,
Headers: s.headers,
})
if err != nil {
return nil, err
}
// Return operation response
var ret CatShardsResponse
if err := s.client.decoder.Decode(res.Body, &ret); err != nil {
return nil, err
}
return ret, nil
}
// -- Result of a get request.
// CatShardsResponse is the outcome of CatShardsService.Do.
type CatShardsResponse []CatShardsResponseRow
// CatShardsResponseRow specifies the data returned for one index
// of a CatShardsResponse. Notice that not all of these fields might
// be filled; that depends on the number of columns chose in the
// request (see CatShardsService.Columns).
type CatShardsResponseRow struct {
Index string `json:"index"` // index name
UUID string `json:"uuid"` // index uuid
Shard int `json:"shard,string"` // shard number, e.g. 1
Prirep string `json:"prirep"` // "r" for replica, "p" for primary
State string `json:"state"` // STARTED, INITIALIZING, RELOCATING, or UNASSIGNED
Docs int64 `json:"docs,string"` // number of documents, e.g. 142847
Store string `json:"store"` // size, e.g. "40mb"
IP string `json:"ip"` // IP address
ID string `json:"id"`
Node string `json:"node"` // Node name
SyncID string `json:"sync_id"`
UnassignedReason string `json:"unassigned.reason"`
UnassignedAt string `json:"unassigned.at"`
UnassignedFor string `json:"unassigned.for"`
UnassignedDetails string `json:"unassigned.details"`
RecoverysourceType string `json:"recoverysource.type"`
CompletionSize string `json:"completion.size"` // size of completion on primaries & replicas
FielddataMemorySize string `json:"fielddata.memory_size"` // used fielddata cache on primaries & replicas
FielddataEvictions int `json:"fielddata.evictions,string"` // fielddata evictions on primaries & replicas
QueryCacheMemorySize string `json:"query_cache.memory_size"` // used query cache on primaries & replicas
QueryCacheEvictions int `json:"query_cache.evictions,string"` // query cache evictions on primaries & replicas
FlushTotal int `json:"flush.total,string"` // number of flushes on primaries & replicas
FlushTotalTime string `json:"flush.total_time"` // time spent in flush on primaries & replicas
GetCurrent int `json:"get.current,string"` // number of current get ops on primaries & replicas
GetTime string `json:"get.time"` // time spent in get on primaries & replicas
GetTotal int `json:"get.total,string"` // number of get ops on primaries & replicas
GetExistsTime string `json:"get.exists_time"` // time spent in successful gets on primaries & replicas
GetExistsTotal int `json:"get.exists_total,string"` // number of successful gets on primaries & replicas
GetMissingTime string `json:"get.missing_time"` // time spent in failed gets on primaries & replicas
GetMissingTotal int `json:"get.missing_total,string"` // number of failed gets on primaries & replicas
IndexingDeleteCurrent int `json:"indexing.delete_current,string"` // number of current deletions on primaries & replicas
IndexingDeleteTime string `json:"indexing.delete_time"` // time spent in deletions on primaries & replicas
IndexingDeleteTotal int `json:"indexing.delete_total,string"` // number of delete ops on primaries & replicas
IndexingIndexCurrent int `json:"indexing.index_current,string"` // number of current indexing on primaries & replicas
IndexingIndexTime string `json:"indexing.index_time"` // time spent in indexing on primaries & replicas
IndexingIndexTotal int `json:"indexing.index_total,string"` // number of index ops on primaries & replicas
IndexingIndexFailed int `json:"indexing.index_failed,string"` // number of failed indexing ops on primaries & replicas
MergesCurrent int `json:"merges.current,string"` // number of current merges on primaries & replicas
MergesCurrentDocs int `json:"merges.current_docs,string"` // number of current merging docs on primaries & replicas
MergesCurrentSize string `json:"merges.current_size"` // size of current merges on primaries & replicas
MergesTotal int `json:"merges.total,string"` // number of completed merge ops on primaries & replicas
MergesTotalDocs int `json:"merges.total_docs,string"` // docs merged on primaries & replicas
MergesTotalSize string `json:"merges.total_size"` // size merged on primaries & replicas
MergesTotalTime string `json:"merges.total_time"` // time spent in merges on primaries & replicas
RefreshTotal int `json:"refresh.total,string"` // total refreshes on primaries & replicas
RefreshExternalTotal int `json:"refresh.external_total,string"` // total external refreshes on primaries & replicas
RefreshTime string `json:"refresh.time"` // time spent in refreshes on primaries & replicas
RefreshExternalTime string `json:"refresh.external_time"` // external time spent in refreshes on primaries & replicas
RefreshListeners int `json:"refresh.listeners,string"` // number of pending refresh listeners on primaries & replicas
SearchFetchCurrent int `json:"search.fetch_current,string"` // current fetch phase ops on primaries & replicas
SearchFetchTime string `json:"search.fetch_time"` // time spent in fetch phase on primaries & replicas
SearchFetchTotal int `json:"search.fetch_total,string"` // total fetch ops on primaries & replicas
SearchOpenContexts int `json:"search.open_contexts,string"` // open search contexts on primaries & replicas
SearchQueryCurrent int `json:"search.query_current,string"` // current query phase ops on primaries & replicas
SearchQueryTime string `json:"search.query_time"` // time spent in query phase on primaries & replicas, e.g. "0s"
SearchQueryTotal int `json:"search.query_total,string"` // total query phase ops on primaries & replicas
SearchScrollCurrent int `json:"search.scroll_current,string"` // open scroll contexts on primaries & replicas
SearchScrollTime string `json:"search.scroll_time"` // time scroll contexts held open on primaries & replicas, e.g. "0s"
SearchScrollTotal int `json:"search.scroll_total,string"` // completed scroll contexts on primaries & replicas
SearchThrottled bool `json:"search.throttled,string"` // indicates if the index is search throttled
SegmentsCount int `json:"segments.count,string"` // number of segments on primaries & replicas
SegmentsMemory string `json:"segments.memory"` // memory used by segments on primaries & replicas, e.g. "1.3kb"
SegmentsIndexWriterMemory string `json:"segments.index_writer_memory"` // memory used by index writer on primaries & replicas, e.g. "0b"
SegmentsVersionMapMemory string `json:"segments.version_map_memory"` // memory used by version map on primaries & replicas, e.g. "0b"
SegmentsFixedBitsetMemory string `json:"segments.fixed_bitset_memory"` // memory used by fixed bit sets for nested object field types and type filters for types referred in _parent fields on primaries & replicas, e.g. "0b"
SeqNoMax int `json:"seq_no.max,string"`
SeqNoLocalCheckpoint int `json:"seq_no.local_checkpoint,string"`
SeqNoGlobalCheckpoint int `json:"seq_no.global_checkpoint,string"`
WarmerCurrent int `json:"warmer.current,string"` // current warmer ops on primaries & replicas
WarmerTotal int `json:"warmer.total,string"` // total warmer ops on primaries & replicas
WarmerTotalTime string `json:"warmer.total_time"` // time spent in warmers on primaries & replicas, e.g. "47s"
PathData string `json:"path.data"`
PathState string `json:"path.state"`
}
// catShardsResponseRowAliasesMap holds the global map for columns aliases
// the map is used by CatShardsService.buildURL.
// For backwards compatibility some fields are able to have the same aliases
// that means that one alias can be translated to different columns (from different elastic versions)
// example for understanding: rto -> RefreshTotal, RefreshExternalTotal
var catShardsResponseRowAliasesMap = map[string]string{
"sync_id": "sync_id",
"ur": "unassigned.reason",
"ua": "unassigned.at",
"uf": "unassigned.for",
"ud": "unassigned.details",
"rs": "recoverysource.type",
"cs": "completion.size",
"fm": "fielddata.memory_size",
"fe": "fielddata.evictions",
"qcm": "query_cache.memory_size",
"qce": "query_cache.evictions",
"ft": "flush.total",
"ftt": "flush.total_time",
"gc": "get.current",
"gti": "get.time",
"gto": "get.total",
"geti": "get.exists_time",
"geto": "get.exists_total",
"gmti": "get.missing_time",
"gmto": "get.missing_total",
"idc": "indexing.delete_current",
"idti": "indexing.delete_time",
"idto": "indexing.delete_total",
"iic": "indexing.index_current",
"iiti": "indexing.index_time",
"iito": "indexing.index_total",
"iif": "indexing.index_failed",
"mc": "merges.current",
"mcd": "merges.current_docs",
"mcs": "merges.current_size",
"mt": "merges.total",
"mtd": "merges.total_docs",
"mts": "merges.total_size",
"mtt": "merges.total_time",
"rto": "refresh.total",
"rti": "refresh.time",
// "rto": "refresh.external_total",
// "rti": "refresh.external_time",
"rli": "refresh.listeners",
"sfc": "search.fetch_current",
"sfti": "search.fetch_time",
"sfto": "search.fetch_total",
"so": "search.open_contexts",
"sqc": "search.query_current",
"sqti": "search.query_time",
"sqto": "search.query_total",
"scc": "search.scroll_current",
"scti": "search.scroll_time",
"scto": "search.scroll_total",
"sc": "segments.count",
"sm": "segments.memory",
"siwm": "segments.index_writer_memory",
"svmm": "segments.version_map_memory",
"sfbm": "segments.fixed_bitset_memory",
"sqm": "seq_no.max",
"sql": "seq_no.local_checkpoint",
"sqg": "seq_no.global_checkpoint",
"wc": "warmer.current",
"wto": "warmer.total",
"wtt": "warmer.total_time",
}