README
¶
istorage
- Simple key-value persistency interface
- Idea borrowed from Project Kaiser storage
Channel: pointer or value
Seems chan RecordParts is more effective than chan *RecordParts
https://groups.google.com/forum/#!topic/golang-nuts/eM_a09l8yU0 You may be surprised how large a struct can get before passing it has a noticeable performance impact compared to passing a pointer to that struct (not to mention accessing data behind a pointer involves an indirection, and especially when the data is shared across processors, there can be additional cost.
BenchmarkChanInterface-4 3000000 428 ns/op 64 B/op 1 allocs/op
BenchmarkChanValue-4 5000000 297 ns/op 0 B/op 0 allocs/op
BenchmarkChanValueFromPointer-4 5000000 301 ns/op 0 B/op 0 allocs/op
BenchmarkChanPointer-4 5000000 395 ns/op 64 B/op 1 allocs/op
Slice or Map
Slice is faster for small sizes https://www.darkcoding.net/software/go-slice-search-vs-map-lookup/
Documentation
¶
Index ¶
Constants ¶
const ( // Aka BO DomainConf = iota // Aka POS DomainOps )
Domain values
const DefaultPart = 0
DefaultPart denotes default part of the record
Variables ¶
var ErrFilteringByOnlyPartsIsProhibited = errors.New("Filtering by only parts is prohibited")
ErrFilteringByOnlyPartsIsProhibited s.e.
var Get func(ctx context.Context, workspaceID int64, recordType int32, IDs []int64, partTypes []int32) (res <-chan Record, perr *error)
Get returns channel for records with given workspaceID and recordType - res buffer length must be zero - res must be sorted by ID, PartType - If IDs is not empty result is filtered by IDs - If IDs is not empty and partTypes is not empty result is filtered by IDs and partTypes - Get must analyze ctx.Err AFTER each write to channel - *perr will be valid when chan is closed - `Get` is intended for fast reading, Cassandra implementation must use CL=1 - ErrFilteringByOnlyPartsIsProhibited must be returned if len(partTypes) > 0 && len(IDs) == 0
var GetEvents func(ctx context.Context, LogID int64, offset1From, offset1To int64) (res <-chan Event, perr *error)
GetEvents between offset1From and offset1To (offset1To INCLUDED into result) Must analyze ctx.Err AFTER each write to channel
Put puts records in transaction
PutEvents batch of events Transaction is not needed (so only some records can be written) Offset1 MUST monotonically increase If it is not GetEvents may stop on first missed number
Functions ¶
Types ¶
type Event ¶ added in v0.12.0
type Event struct {
Offset1 int64
Offset2 int64
WSID int64
Domain int32
Type int32
OriginDeviceID int64
OriginTimestamp int64
OriginOffset int64
MyTimestamp int64
Data []byte
MetaData []byte
}
Event s.e.
type Record ¶
type Record struct {
RecordType int32
ID int64
DocType int32
DocID int64
ParentType int32
ParentID int64
PartType int32
PartID int64
// Offset of the event which affected the value
Offset int64
// 0-deleted, 1-active, 2-closed
State int32
// Data
Value []byte
}
Record used by Put
Source Files