* add an Aerospike storage backend * go mod vendor * add Aerospike storage configuration docs * review fixes * bump aerospike client to v3.1.1 * rename the defaultHostname variable * relocate the docs page
50 KiB
Change History
November 9 2020: v3.1.1
Hotfix release. We recommend upgrading to this version, or cherry-picking the changeset to your vendored version if possible.
-
Fixes
- Handle cleanup cases in
Offer
andDropIdleTail
forsingleConnectionHeap
. (Github #318) - Unlock the mutex in
singleConnectionHeap.Poll
if called after cleanup. (Github #323) thanks to linchuan4028
- Handle cleanup cases in
September 10 2020: v3.1.0
Minor fix release.
-
Fixes
- Fixes an issue where initial tend was not adhering to the
ClientPolicy.Timeout
. (CLIENT-1344)
- Fixes an issue where initial tend was not adhering to the
August 19 2020: v3.0.5
Minor fix release.
-
Fixes
- Corrects the maximum bin name size in error message.
- Fixes geo coordinates in predexp tests due to more strict server validation.
- Fixes misspelled words and doc links. PR #311, thanks to Abhay
July 7 2020: v3.0.4
Minor fix release.
-
Fixes
- Fixes
Client.SetWhitelist
API. - Fixes an issue where
Whilelist
was not set duringQueryRole
.
- Fixes
July 3 2020: v3.0.3
Minor fix release.
-
Fixes
- Resolves an issue where batch retry could return some nil records in some situations.
June 24 2020: v3.0.2
Minor improvement release.
-
Fixes
- Fixes an issue where if a slice was pre-assigned on a struct, the data would not be allocated to it in reflection API. PR #302, thanks to gdm85
- Fixes an issue where
Node.GetConnection()
could in rare circumstances return no connection without anerror
. This would potentially cause a panic in VERY slow production servers.
-
Improvements
- Converts a few panics to errors in wire protocol encoding/decoding. Resolves issue #304.
June 17 2020: v3.0.1
Minor bug fix release.
-
Fixes
- Fixes caching of embedded structs with options in alias. Resolves issue #301.
June 8 2020: v3.0.0
Major feature release. There are a few minor breaking API changes. See ClientPolicy
.
Note: There has been significant changes to clustering code. We recommend extensive testing before using in production.
-
New Features
- Adds support for Relaxed Strong Consistency mode.
- Adds support for whitelists in Roles.
May 28 2020: v2.12.0
Minor feature release.
-
New Features
- Adds
MapCreateOp
andListCreateOp
inContext
for CDTs.
- Adds
May 27 2020: v2.11.0
Major feature release.
-
New Features
- Adds HyperLogLog support.
-
Improvements
- Exports
estimateSize
onValue
Datastructure. PR #299, thanks to Sainadh Devireddy - Adds more detail regarding
ClientPolicy.IdleTimeout
in the documentation, and changes the default value to 55 seconds.
- Exports
May 12 2020: v2.10.0
Minor feature release.
-
New Features
- Adds
ClientPolicy.MinConnectionsPerNode
.
- Adds
-
Improvements
- Returns distinct error when empty slice is passed to BatchGetObjects. PR #297, thanks to Mohamed Osama
March 14 2020: v2.9.0
Minor feature release.
-
New Features
- Supports use of anonymous structs in reflection API. PR #287, thanks to 小马哥
March 4 2020: v2.8.2
Hotfix.
-
Fixes
- Fixes a race condition introduced in the last release.
March 4 2020: v2.8.1
Minor fix and improvements.
-
Improvements
- Uses a
sync.Pool
to preserve the connection buffers on close to remove pressure from the allocator and the GC during connection churns.
- Uses a
-
Fixes
- Cleanup the data structure cross refs on Cluster.Close to help GC free the objects.
February 28 2020: v2.8.0
Minor feature release.
-
New Features
- Allows
,omitempty
tag to be used in struct tags. It behaves the same as the stdlib json. Note that there should be no whitespace between the comma and the rest of the tag.
- Allows
January 30 2020: v2.7.2
Minor fix release.
-
Fixes
- Resolves an issue where an invalid/malformed compressed header could cause a panic by reporting wrong compressed data size. Resolves #280.
January 20 2020: v2.7.1
Minor fix release.
-
Fixes
- Fixes an issue where an error was not checked after a read in
multi_command.go
and would cause a panic. Resolves #280.
- Fixes an issue where an error was not checked after a read in
December 24 2019: v2.7.0
Minor feature and fix release.
-
New Features
- Adds support for client/server wire transport compression.
- Adds support for descending CDT list order.
-
Fixes
- Fixes an issue where unpacking
Value
objects would cause an infinite loop. PR #273, thanks to small-egg
- Fixes an issue where unpacking
November 25 2019: v2.6.0
Minor feature release.
-
New Features
- Supports correct Query/Scans via
Scan/QueryPolicy.FailOnClusterChange
- Supports correct Query/Scans via
-
Fixes
- Fixes an issue where the client using multiple seeds via DNS or Load Balancer would fail to connect if more than one of them were unreachable.
November 8 2019: v2.5.0
Major feature release.
-
New Features
- Adds support for predicate expressions in all transactions. See
Policy.Predexp
.
- Adds support for predicate expressions in all transactions. See
October 29 2019: v2.4.0
Major feature release.
-
New Features
- Adds support for bitwise operations.
- Adds support for nested CDTs.
October 17 2019: v2.3.0
Major feature release.
-
New Features
- Adds support for mixed security modes in cluster to enable rolling upgrade with security changes.
- Adds support for delete record operation
DeleteOp()
inClient.Operate()
. - Adds support for write operations in background scan/query.
- Adds support for
Scan/QueryPolicy.RecordsPerSecond
field to limit throughput.
August 13 2019: v2.2.1
Minor improvement release.
-
Improvements
- Supports the
Write
role in server v4.6.0.2+
- Supports the
May 21 2019: v2.2.0
Minor Fixes and improvements release.
-
Fixes
- Fixes an issue where an empty connection pool would cause a lock contention that would in turn lead to high CPU load.
- Fixes an issue where in some circumstances connection pool would be depleted of connections.
- Fixes an issue where the replica node would not be selected in case of master-node failure in
Policy.ReplicaPolicy.SEQUENCE
for reads.
-
Improvements
- Transactions will not count a lack of connection in the node's connection pool as an iteration anymore.
April 25 2019: v2.1.1
Minor Fixes and improvements release.
-
Fixes
- Fixes an issue where meta tags were ignored in reflection API for
ScanAllObjects
/QueryObjects
/BatchGetObjects
. Resolves #260.
- Fixes an issue where meta tags were ignored in reflection API for
-
Improvements
- Tend won't send
rack:
command to the nodes ifClientPolicy.RackAware
is not set. PR #259, thanks to Dmitry Maksimov - Adds a new GeoJson example.
- Tend won't send
April 11 2019: v2.1.0
Minor Feature and Improvements release.
-
New Features
- Adds
WarmUp
method forClient
,Cluster
andNode
. This method will fill the connection queue to ensure maximum and smooth performance on start up.
- Adds
-
Improvements
- Simplify connection Timeout calculation and floor the min timeout to 1ms.
- Simplify resetting server timeout for each iteration.
- Adds a few pre-defined errors to avoid allocating them during runtime.
-
Changes
- Adds a TLS connection example.
- Adds
Cap
method toconnectionHeap
.
March 19 2019: v2.0.0
Major release. There are some breaking changes, both syntactically and semantically.
Most changes are minor, and can be fixed with relative ease.
The only major issue is that the behavior of the client when a key does not exist has changed.
It used to return no error, but nil
Record.Bins
. Now it returns ErrKeyNotFound
error.
This is a significant changes, and you should search your code for all instances of Bins == nil
and adapt the code accordingly.
-
Major:
- Optimizes connection creation out of the transaction pipeline and makes it async.
- Put a threshold on the number of connections allowed to open simultaneously. Controlled via
ClientPolicy.OpeningConnectionThreshold
. - Do not clear partition map entry when a node reports that it no longer owns that partition entry.
- Uses rolling timeout instead of strict timeout for connections.
- Remove
ToValueArray
andToValueSlice
methods to discourage such suboptimal use. ChangesQueryAggregate
signature to remove the need for those methods. - Remove unnecessary conversion from BinMap to Bins in reflection API to speedup the command an avoid unnecessary memory allocations.
- Use shorter intervals with exponential back-off for tasks.
-
Breaking:
Get
/Put
/Touch
/Operate
andExecuteUDF
commands will return anErrKeyNotFound
error when the key does not exist in the database. The old behavior used to be not to return an error, but have an emptyRecord.Bins
.- Renames
Statement.Addfilter
toStatement.SetFilter
, change the name and type ofStatement.Filters
toStatement.Filter
. - Remove
ClientPolicy.RequestProleReplicas
. THe client will always request them. - Removes
ScanPolicy.ServerSocketTimeout
andQueryPolicy.ServerSocketTimeout
in favor of the already existingPolicy.SocketTimeout
. - Renames
Policy.Timeout
toPolicy.TotalTimeout
to make the naming consistent with other clients. - Moves
atomic
package to internal. - Moves
ParticleType
package to internal. - Moves
RequestNodeInfo
andRequestNodeStats
to methods on Node object, and addsInfoPolicy
to the relevant API signatures. - Removes
WaitUntilMigrationIsFinished
fromScan
/Query
policies. - Changes
NewConnection
method signature, makesLoginCommand
private. - Makes
OperationType
private. - Remove long deprecated method for pool management.
- Removes unused
ReadN
method inConnection
. - Embeds Policies as values and not pointers inside
MultiPolicy
,ScanPolicy
,QueryPolicy
-
Minor:
- Fixes a race condition in the
AdminCommand
. - Synchronize the
XORShift
to avoid race conditions. - Completely removes deprecated LDT code.
- Fixes a race condition in the
March 11 2019: v1.39.0
Major improvements Release.
-
Improvements
- Significantly improves
Batch
/Scan
/Query
/UDF
/QueryAggregate
performance, up to 10x depending on the number of records.
- Significantly improves
-
Changes
- Removes
BatchPolicy.UseBatchDirect
from the code since it is not supported on the server anymore.
- Removes
February 21 2019: v1.38.0
-
New Features
- Support new server
truncate-namespace
command viaClient.Truncate
whenset
is not specified.
- Support new server
-
Improvements
- The client will not clear a partition map entry when a node reports that it no longer owns that partition entry until another node claims ownership.
- Adapt UDF test for new server changes. The server will not return an error after
RemoveUDF
if the UDF did not exist. - Improves a few tests and relaxes tolerances in tests to accommodate slower cloud test environments.
-
Fixes
- Fixes a race condition in XOR shift RNG.
- Fixes a race condition in the AdminCommand.
December 3 2018: v1.37.0
-
New Features
- Support lut-now parameter for Client.Truncate() in servers that support and require it.
- Added support for CDT Map Relative Ops:
MapGetByKeyRelativeIndexRangeOp
,MapGetByKeyRelativeIndexRangeCountOp
,MapGetByValueRelativeRankRangeOp
,MapGetByValueRelativeRankRangeCountOp
,MapRemoveByKeyRelativeIndexRangeOp
,MapRemoveByKeyRelativeIndexRangeCountOp
. - Added support for CDT List Relative Ops:
ListGetByValueRelativeRankRangeOp
,ListGetByValueRelativeRankRangeCountOp
,ListRemoveByValueRelativeRankRangeOp
,ListRemoveByValueRelativeRankRangeCountOp
. - Added
INFINITY
andWILDCARD
values for use in CDT map/list comparators.
-
Improvements
- Increase default
Policy.SocketTimeout
to 30s. IfSocketTimeout
is longer thanTimeout
,Timeout
will be used instead silently. This change is done for the client to perform more intuitively in cloud environments. - Never return a random node if a node was not found in the partition map.
- Return more descriptive error messages on various partition map and other node related errors.
- Increase default
-
Changes
- Remove the ability to force old batch direct protocol on the client because the server will be removing support for the old batch direct protocol.
- Update admin message version to 2.
- Remove unused error codes.
- Remove Go 1.7 and 1.8 from travis tests due to incompatibility with the test framework.
November 1 2018: v1.36.0
Feature Release.
-
New Features
- Support rackaware feature. You need to set the
ClientPolicy.RackAware = true
, and set theClientPolicy.RackId
. All read operations will try to choose a node on the same rack ifPolicy.ReplicaPolicy = PREFER_RACK
. This feature is especially useful when the app/cluster are on the cloud and network throughput over different zones are price differently.
- Support rackaware feature. You need to set the
-
Improvements
- Update Operate command documentation.
- Improve an expectation in a CDT Map test.
- Move UDF object test to the proper file.
- Support float64 struct fields when the value of the field has been changed inside lua and set to int - will only affect clusters which support float.
- Fixes an issue where key value was sent and cause server PARAMETER_ERROR via the operate command if policy.SendKey was set but no write operations were passed.
- Updated README example with clarification.
-
Fixes
- Fixes an issue where multiple operation results for a bin would be appended to the first result if it was a list.
October 2 2018: v1.35.2
Improvement release.
-
Improvements
- Do not allocate a partition map on each tend unless needed.
- Adds
ConnectionsClosed
stat and sets the connection and dataBuffer to nil in a few places to help the GC. - Use a heap data structure for connection pooling instead of a queue. This allows better management of connections after a surge, since it keeps the unused connection in the bottom of the heap to close. It also helps with performance a bit due to better caching of the data structure in CPU.
September 18 2018: v1.35.1
Hot fix release. We recommend updating to this version if you are using authentication.
-
Fixes
- Fixes a regression to avoid hashing passwords per each login, using the cached password.
-
Changes
- Minor code clean up and dead code removal.
August 29 2018: v1.35.0
-
New Features
- Support for external authentication (LDAP).
- Support Map and List WriteFlags:
NoFail
andPartial
. - Support load balancers as seed node.
-
Changes
- Change default Scan/Query
ServerSocketTimeout
to 30s.
- Change default Scan/Query
-
Improvements
- Adds
QueryPolicy.ServerSocketTimeout
andQueryPolicy.FailOnClusterChange
for when the queries are automatically converted to scans. - Minor documentation improvements.
- Synchronize logging at all situations.
- Add -debug switch to allow logging at debug level in tests.
- Allow the user to define the namespace for tests to run on.
- Adds
-
Fixes
- Fix a few go vet errors for Go 1.11.
- Fixes minor unsigned length conversions for admin command.
August 29 2018: v1.34.2
Fix release.
-
Fixes
- Use pointer receiver for
AerospikeError.SetInDoubt
andAerospikeError.MarkInDoubt
. - Remove unused variable in truncate test.
- Use pointer receiver for
-
Changes
- Add Go 1.11 to Travis' test versions.
- Use the last error code in MaxRetries timeout errors for Go 1.11.
August 9 2018: v1.34.1
Hot fix release. We recommend updating to this version asap, especially if you are using the Strong Consistency feature.
-
Fixes
- Fixes an issue where a race condition was preventing the partition table to form correctly. (CLIENT-1028)
July 17 2018: v1.34.0
-
Changes
- Removed the LDT code completely.
- Adds build tag
app_engine
for compatibility with Google's App Engine. Query Aggregate features are not available in this mode due to lua limitations.
-
Improvements
- Document how to use AerospikeError type in the code.
- Allow Task.OnComplete() to be listened to by multiple goroutines. Thanks to HArmen
-
Fixes
- Fixes an issue where
ClientPolicy.FailIfNotConnected
flag was not respected. - Fixes a merging issue for PartitionMap, and add a naive validation for partition maps. (CLIENT-1027)
- Fixes an issue where
June 11 2018: v1.33.0
-
New Features
- Adds
BatchPolicy.AllowPartialResults
flag to allow the batch command return partial records returned from the cluster. - Adds
INVERTED
flag to theMapReturnType
. Take a look at INVERTED test incdt_map_test.go
to see how to use it. - Adds a lot of new Ordered Map and List operations and brings the client up to date with the latest server API.
- Adds
-
Changes
- Use the default values for
BasePolicy
inBatchPolicy
to keep the behavior consistent with the older releases.
- Use the default values for
-
Improvements
- Adds a recover to the tend goroutine to guarantee the client will recover from internal crashes.
- Removes unneeded type casts.
- Uses the new stat name for migrations check.
-
Fixes
- Fixes TTL in
GetObject
andBatchGetObject
reflection API. - Handle extension marker in List headers.
- Fixes TTL in
March 15 2018: v1.32.0
Major feature release.
-
New Features
- Support for Strong Consistency mode in Aerospike Server v4. You will need to set the
policy.LinearizeRead
totrue
. AddsAerospikeError.InDoubt()
method. - Set the resulting
Record.Key
value in Batch Operations to the Original User-Provided Key to preserve the original namespace/set/userValue and avoid memory allocation.
- Support for Strong Consistency mode in Aerospike Server v4. You will need to set the
-
Changes
- Does not retry on writes by default, and put a 100ms timeout on all transactions by default.
- Changed some warn logs to debug level.
- Add missing stats counters to improve statistics reports.
- Uses sync.Once instead of sync.Mutex for
Connection.close
method. - Added
DefaultBufferSize
for initial buffer size for connections.
-
Fixes
- Fix the tests for object marshalling to account for monotonic time in Go v1.8+.
- Stops the ongoing tends on initial connection errors.
November 29 2017: v1.31.0
Feature release.
-
New Features
- Support for newer Batch Protocol. Add
BatchGetComplex
for complex batch queries. Old batch API upgraded to automatically support the new protocol under the hood, unlessBatchPolicy.UseBatchDirect
flag is set totrue
.
- Support for newer Batch Protocol. Add
-
Changes
- Renames ResultCode
NO_XDS
toALWAYS_FORBIDDEN
. - Makes
SERVER_NOT_AVAILABLE
a client generated error.
- Renames ResultCode
October 12 2017: v1.30.0
Fix and improvements release.
-
Changes
- Deprecated LDTs and removed them from the official build.
- Change supported go versions to 1.7+ due to gopher-lua requiring
Context
.
-
Improvements
- Get socket timeout once per command execution, do not redefine err var in command execution loop. PR #211, thanks to Maxim Krasilnikov
- Allow running a UDF on a specific node only.
- Close cluster only once. PR #208, thanks to Jun Kimura
- Use actual cluster name in tests instead of assuming
null
. - Check for the type of error as well in Duplicate Index Creation.
- Update description for error code 21. PR #207, thanks to Maxim Krasilnikov
September 5 2017: v1.29.0
Feature and improvements release.
-
New Features
- Added
ListIncrementOp
to the CDT list operations. - Added
SEQUENCE
to replica policies.
- Added
-
Improvements
- Tweaked node removal algorithm to cover more corner cases.
- Make
predExp
interface public. Closes issue #205. - Added more stats to the
Client.Stats()
.
August 17 2017: v1.28.0
Feature, Performance improvements and bug fix release.
-
New Features
- Added
Client.Stats()
method to get client's internal statistics. - Added
Policy.SocketTimeout
to differentiate between network timeouts and the total transaction timeouts. - Support
policy.IncludeBinData
for queries. Only for servers that support this feature. - Minor documentation updates.
- Return key not found exception (instead of returning nil record) for Operate() command where operations include a write.
- Added
-
Improvements
- Close the tend connection when closing node connections.
- Added Connection finalizer to make sure all connections are closed eventually.
- Automatically retry failed info requests on async tasks before returning an error.
- Updated build instructions for the benchmark tool.
- Make digest_modulo test deterministic.
- Relax predexp_modulo test a bit to avoid occasional failures.
-
Fixes
- Indirect CAS ops to prevent the compiler from optimizing them out.
- Return errors instead of nil.
April 25 2017: v1.27.0
Feature, Performance improvements and bug fix release.
-
New Features
- Added
BatchGetObjects
method. - Added Exponential Backoff by introducing
BasePolicy.SleepMultiplier
. Only Values > 1.0 are effective. PR #192, thanks to Venil Noronha
- Added
-
Improvements
- Packer tries to see if it can use generic data types before using reflection.
- Operations, including CDTs do not allocate a buffer anymore, unless reused.
-
Incompatible changes:
BinName
andBinValue
are not exported inOperation
anymore. These fields shouldn't have been used anyway sinceOperation
s used to cache their internal command.
-
Fixes
- Documentation Fixes. Thanks to Nassor Paulino da Silva and HArmen
April 5 2017: v1.26.0
Feature, Performance improvements and bug fix release.
-
New Features
- Predicate API is supported (for server v3.12+)
- Added
Truncate
method to quickly remove all data from namespaces or sets (for server v3.12+). - Support
ScanPolicy.ServerSocketTimeout
(for server v3.12+). - Support
ClientPolicy.IgnoreOtherSubnetAliases
to ignore hosts from other subnets. PR #182, thanks to wedi-dev
-
Improvements
- Added a lot of predefined generic slice and map types in
NewValue
method to avoid hitting reflection as much as possible. - Fix
go vet
complaints.
- Added a lot of predefined generic slice and map types in
-
Fixes
- Allow streaming commands (scan/query/aggregation) to retry unless the error occurs during parsing of the results. Fixes issue #187
- Use
net.JoinHostPort
to concatinate host and port values instead of doing it directly. Fixes some issues in IPv6 connection strings. - Improved initial Tend run.
- Fixes
cluster-name
checking bug.
March 8 2017: v1.25.1
Hot fix release. Updating the client is recommended.
-
Fixes
- Fixed an issue where errors in Scan/Query unmarshalling would be duplicated and could cause a deadlock.
February 28 2017: v1.25.0
Performance improvements and fix release.
-
Improvements
- Check tend duration and compare it to tend interval, and warn the user if tend takes longer than tend interval.
- Seed the cluster concurrently, and return as soon as any of the seeds is validated.
- Tend the cluster concurrently. Allows use of very big clusters with no delay.
- Partitions the connection queue to avoid contention.
- Cluster partition map is merged from all node fragments and updated only once per tend to reduce contention to absolute minimum.
-
Fixes
- Fixed an issue where a valid but unreachable seed could timeout and stall connecting and tending the cluster..
- Fix result code comments.
January 11 2017: v1.24.0
Minor feature and fix release.
-
New Features
- TLS/SSL connections are now officially supported.
- Added Role/Privilege API.
-
Improvements
- Return a client-side error when no ops are passed to the operate command.
- Export error attribute in
NodeError
- Do not attempt to refresh peers if it is not supported by the nodes.
-
Fixes
- Use namespace default-ttl for tests instead of assuming 30d
- Always drain scan connections after parsing the records.
- Fix panic in GetObject() if all bins in result is nil. PR #172, thanks to Hamper
- Fix WritePolicy usage with UDF. PR #174, thanks to Bertrand Paquet
- Close connection right when it has an io error and don't wait for the caller.
December 20 2016 : v1.23.0
Minor feature and fix release.
-
New Features
- Exposes the internal
client.Cluster
object to the users. - Added New API for high-performance complex data type packing, and removed the old API.
- Exposes the internal
-
Improvements
- Only update the partition map if the partition generatio has changed.
- Use tend connection for user management commands.
- Marks LargeList as deprecated. Use CDT methods instead.
- Always validate the message header to avoid reading the remainder of other command buffers.
- Removes GeoJson from key helper.
- Improves tend algorthm to allow complete disconnection from the cluster if none of the clusters are accessible.
PutObject
method will now accept objects as well. PR #156, thanks to Sarath S Pillai
-
Fixes
- Do not attemp to add a node which were unaccessible to avoid panic.
- Fix invalid connectionCount. PR #168, thanks to Jun Kimura
- Fixes minor bug that didn't return the error on reading from the connection during scans.
November 29 2016 : v1.22.0
Hot fix release. Please upgrade if you have been using other aerospike clients with your database parallel to Go.
-
Fixes
- Fixes an issue where short strings in Lists and Maps wouldn't unpack correctly. Resolves #161.
November 16 2016 : v1.21.0
Minor fix release.
-
New Features
- Added new constants for expiration in
WritePolicy
:TTLServerDefault
,TTLDontExpire
,TTLDontUpdate
- Added new constants for expiration in
-
Improvements
- Corrects typos in the code. PR #142, thanks to Muyiwa Olurin
- Use the tend connection for
RequestInfo
commands.
-
Fixes
- Fixes an issue where TTL values were calcualted wrongly when they were set not to expire.
- Fixes an issue where
PutObjects
would marshal[]byte
toList
in database. PR #152, thanks to blide - Fixes an issue where
Recordset
could leak goroutines. PR #153, thanks to Deepak Prabhakara
October 25 2016 : v1.20.0
Major improvements release. There has been major changes in the library. Please test rigorously before upgrading to the new version.
-
New Features
-
Let user define the desired tag for bin names in structs using
SetAerospikeTag
function. -
Added
as_performance
build tag to avoid including the slow convenience API which uses reflections in the client code. To use this feature, you should include -tags="as_performance" when building your project.NOTICE: Keep in mind that your code may not compile using this flag. That is by design.
-
-
Improvements
- Added special packer for map[string]interface{} in
NewValue
method. - Avoid allocating memory for Map and List values.
- Allocate commands on the stack to avoid heap allcations.
- Avoid allocating memory for
packer
. - Avoid Allocating memory in computeHash for keys.
- Avoid allocating memory in Ripe160MD digest.
- Removed BufferPool and moved buffers to
Connection
objects to remove lock contention. - Added
ListIter
andMapIter
interfaces to support passing Maps and Lists to the client without using reflection.
- Added special packer for map[string]interface{} in
October 14 2016 : v1.19.0
Major feature and improvement release.
-
New Features
-
Support TLS secured connections. (Feature will be supported in coming server releases.)
-
Support IPv6 protocol. Supported by Aerospike Server 3.10+.
-
Support
cluster-name
verification. Supported by Aerospike Server 3.10+. -
Support new peers info protocol. Supported by Aerospike Server 3.10+.
-
-
Improvements
-
Will retry the operation even when reading from the buffer. Set
Policy.MaxRetries = 0
to avoid this behavior. PR #143, thanks to Hector Jusforgues -
Much improved cluster management algorithm. Will now handle the case where multiple nodes go down simultaneously, still protecting against split brain rogue nodes.
-
-
Fixes
-
Try all alias IPs in node validator. Resolves #144.
-
Updated job status check for execute tasks.
-
August 19 2016 : v1.18.0
Minor improvements release.
-
New Features
- Support 'Durable Deletes' for the next version of Aerospike Server Enterprise.
-
Improvements
-
Don't run tests for features that are not supported by the server.
-
Added new server error codes.
-
July 27 2016 : v1.17.1
Minor improvements release.
-
Improvements
-
Add
TaskId()
method forRecordset
. -
Cleanup indexes after test cases.
-
Keep connections on recoverable server errors.
-
Return the error on unexpected keys in
BatchCommandGet/Header
. -
Use the same client object in tests and support using replicas on travis.
-
July 19 2016 : v1.17.0
Major feature and improvement release.
-
New Features
- Client now supports distributing reads from Replicas using
ClientPolicy.RequestProleReplicas
andPolicy.ReplicaPolicy
- Client now supports distributing reads from Replicas using
-
Improvements
-
Cluster.GetConnection
will now retry to acquire a connection until timeout. -
Client.DropIndex
method now blocks until all nodes report the index is dropped. -
Async tasks like
CreateIndex
will retry a few times before deciding a non-existing job means it has finished. -
Don't use math.MaxInt64, it breaks 32-bit builds. PR #139, thanks to Cameron Sparr
-
-
Fixes
-
Maps with 0 elements will automatically shortcut to unordered empty maps.
-
Return the error in BatchCommandGet on parse error.
-
June 28 2016 : v1.16.3
Major bugfix release. Update recommended.
-
Improvements
-
Skip LDT tests if LDT is not enabled.
-
Returns last error after all retry attempts to run a command are exhausted.
-
Reserves a connection for tend operation to avoid dropping a node when high load prevents acquiring a proper connection.
-
Added Finalizers to
Client
andRecordset
. Both will be automatically closed by the GC.
-
-
Fixes
-
Fixes an issue where
services-alternate
wasn't used inNode.addFriends()
when instructed so in the policy. -
Fixes an issue where object metadata wasn't cached if
QueryObjects
was called beforePutObject
. -
Fixes an issue where idle connections were not dropped.
-
Fixes an issue where requested buffer sizes were not guarded against negative numbers.
-
June 7 2016 : v1.16.2
Minor bugfix release.
-
Fixes
- Fixes an issue where empty unordered maps were confused with CDT maps.
June 6 2016 : v1.16.1
Minor bugfix release.
-
Fixes
- Fixes an issue where complex maps and lists weren't unmarshalled correctly in
GetObject
method.
- Fixes an issue where complex maps and lists weren't unmarshalled correctly in
June 2 2016 : v1.16
Major feature and improvements release.
NOTICE: Due to the relatively extensive code overhaul, upgrade with caution.
-
New Features
- Added CDT Ordered Map API. (Requires server v3.8.3+)
-
Improvements
-
Removed mutexes from
Cluster
andNode
code. -
Improved code quality using various linters.
-
May 27 2016 : v1.15
Minor fixes and improvements release.
-
Fixes
- Fixed an issue where unmarshalling embedded structs and pointers didn't work properly if they were tagged.
May 16 2016 : v1.14
Minor fixes and improvements release.
-
Fixes
-
Fixed an issue in which go-routines were leaked in
Results()
method ofRecordset
on cancellation. Based on PR #128, thanks to Noel Cower -
Fixed issues regarding leaked goroutines in
Cluster.WaitTillStablized()
,Cluster.MigrationInProgress()
, andCluster.WaitUntillMigrationIsFinished()
methods. PR #126, thanks to Anton
-
-
Improvements
-
Improved cluster
tend()
logic. -
Added
Recordset.Read()
method. -
Minor fixes in docs and code formatting. Thanks to Andrew Murray and Erik Dubbelboer
-
April 1 2016 : v1.13
Minor features and improvements release.
-
New Features
- Added
NewGeoWithinRegionForCollectionFilter
,NewGeoRegionsContainingPointForCollectionFilter
,NewGeoWithinRadiusForCollectionFilter
for queries on collection bins.
- Added
-
Fixes
- Fixed an issue in which bounded byte arrays were silently being dropped as map keys.
-
Improvements
-
Removed and fixed unused assignments and variables.
-
Fixed typos in the comments.
-
Minor changes and formatting. PR #124, thanks to Harmen
-
March 8 2016 : v1.12
Minor features and improvements release.
-
New Features
- Support Metadata in struct tags to fetch TTL and Generation via
GetObject
. Notice: Metadata attributes in an struct are considered transient, and won't be persisted.
Example:
type SomeStruct struct { TTL uint32 `asm:"ttl"` // record time-to-live in seconds Gen uint32 `asm:"gen"` // record generation A int Self *SomeStruct } key, _ := as.NewKey("ns", "set", value) err := client.PutObject(nil, key, obj) // handle error here rObj := &OtherStruct{} err = client.GetObject(nil, key, rObj)
- GeoJSON support in Lists and Maps
- Support Metadata in struct tags to fetch TTL and Generation via
-
Improvements
-
Use
ClientPolicy.timeout
for connection timeout when refreshing nodes -
Added new server error codes
-
Protect RNG pool against low-precision clocks during init
-
Better error message distingushing between timeout because of reaching deadline and exceeding maximum retries
-
-
Fixes
-
Fixed object mapping cache for anonymous structs. PR #115, thanks to Moshe Revah
-
Fixed an issue where
Execute()
method wasn't observing theSendKey
flag in Policy.
-
February 9 2016 : v1.11
Minor features and improvements release.
-
New Features
-
Can now use
services-alternate
for cluster tend. -
New CDT List API:
ListGetRangeFromOp
,ListRemoveRangeFromOp
,ListPopRangeFromOp
-
-
Improvements
-
Improves marshalling of data types into and out of the Lua library and avoids marshalling values before they are needed.
-
Returns error for having more than one Filter on client-side to avoid confusion.
-
Increases default
ClientPolicy.Timeout
and return a meaningful error message when the client is not fully connected to the cluster afterwaitTillStabilized
call
-
January 13 2016 : v1.10
Major release. Adds Aggregation.
-
New Features
-
Added
client.QueryAggregate
method.-
For examples regarding how to use this feature, look at the examples directory.
-
You can find more documentation regarding the Aggregation Feature on Aerospike Website
-
-
-
Improvements
- Improve Query/Scan performance by reading from the socket in bigger chunks
December 14 2015 : v1.9
Major release. Adds new features.
-
New Features
-
Added CDT List operations.
-
Added
NewGeoWithinRadiusFilter
filter for queries.
-
-
Changes
- Renamed
NewGeoPointsWithinRegionFilter
toNewGeoWithinRegionFilter
- Renamed
December 1 2015 : v1.8
Major release. Adds new features and fixes important bugs.
-
New Features
-
Added
ScanAllObjects
,ScanNodeObjects
,QueryObjects
andQueryNodeObjects
to the client, to facilitate automatic unmarshalling of data similar toGetObject
.- NOTICE: This feature and its API are experimental, and may change in the future. Please test your code throughly, and provide feedback via Github.
-
Added
ScanPolicy.IncludeLDT
option (Usable with yet to be released server v 3.7.0) -
Added
LargeList.Exist
method.
-
-
Improvements
-
Makes Generation and Expiration values consistent for WritePolicy and Record.
- NOTICE! BREAKING CHANGE: Types of
Record.Generation
andRecord.Expiration
, and alsoWritePolicy.Generation
andWritePolicy.Expiration
have changed, and may require casting in older code.
- NOTICE! BREAKING CHANGE: Types of
-
Refactor tools/asinfo to be more idiomatic Go. PR #86, thanks to Tyler Gibbons
-
Many documentation fixes thanks to Charl Matthee and Tyler Gibbons
-
-
Fixes
-
Changed the
KeepConnection
logic from black-list to white-list, to drop all -
Fix RemoveNodesCopy logic error.
-
Add missing send on recordset Error channel. PR #99, thanks to Geert-Johan Riemer
-
Fix skipping of errors/records in (*recordset).Results() select after cancellation. PR #99, thanks to Geert-Johan Riemer
-
October 16 2015 : v1.7
Major release. Adds new features and fixes important bugs.
-
New Features
-
Added support for Geo spatial queries.
-
Added support for creating indexes on List and Map bins, and querying them.
-
Added support for native floating point values.
-
Added
ClientPolicy.IpMap
to use IP translation for alias recognition. PR #81, Thanks to Christopher Guiney
-
-
Improvements
- Cosmetic change to improve code consistency for
PackLong
inpacker.go
. PR #78, Thanks to Erik Dubbelboer
- Cosmetic change to improve code consistency for
-
Fixes
-
Fixes an issue when the info->services string was malformed and caused the client to panic.
-
Fixes an issue with unmarshalling maps of type map[ANY]struct{} into embedded structs.
-
Fixes issue with unmarshalling maps of type map[ANY]struct{} into embedded structs.
-
Fixes an issue with bound checking. PR #85, Thanks to Tait Clarridge
-
Fixes aa few typos in the docs. PR #76, Thanks to Charl Matthee
-
August 2015 : v1.6.5
Minor maintenance release.
-
Improvements
- Export
MaxBufferSize
to allow tweaking of maximum buffer size allowed to read a record. If a record is bigger than this size (e.g: A lot of LDT elements in scan), this setting wil allow to tweak the buffer size.
- Export
July 16 2015 : v1.6.4
Hot fix release.
-
Fixes
- Fix panic when a scan/query fails and the connection is not dropped.
July 9 2015 : v1.6.3
Minor fix release.
-
Improvements
- Improved documentation. PR #64 and #68. Thanks to Geert-Johan Riemer
-
Fixes
-
Fix a bunch of golint notices. PR #69, Thanks to Geert-Johan Riemer
-
Connection.Read() total bytes count on error. PR #71, Thanks to Geert-Johan Riemer
-
Fixed a race condition on objectMappings map. PR #72, Thanks to Geert-Johan Riemer
-
Fixed a few uint -> int convertions.
-
June 11 2015 : v1.6.2
Minor fix release.
-
Improvements
- Improved documentation. Replaced all old API references regarding Recordset/Query/Scan to newer, more elegant API.
-
Fixes
-
Fixed an issue where erroring out on Scan would result a panic.
-
Fixed an issue where
Statement.TaskId
would be negative. convertedStatement.TaskId
touint64
-
June 9 2015 : v1.6.1
Minor fix release.
-
Fixes
-
Fixed an issue where marshaller wouldn't marshal some embedded structs.
-
Fixed an issue where querying/scanning empty sets wouldn't drain the socket before return.
-
May 30 2015 : v1.6.0
There's an important performance regression bug fix in this release. We recommend everyone to upgrade.
-
New Features
-
Added New LargeList API.
- NOTICE! BREAKING CHANGE: New LargeList API on the Go Client uses the New API defined on newer server versions. As Such, it has changed some signatures in LargeList.
-
-
Fixes
-
Fixed an issue where connections where not put back to the pool on some non-critical errors.
-
Fixed an issue where Object Unmarshaller wouldn't extend a slice.
-
Decode RegisterUDF() error message from base64
-
Fixed invalid connection handling on node connections (thanks to @rndive)
-
May 15 2015 : v1.5.2
Hotfix release.
-
Fixes
- Fixed a branch-merge mistake regarding error handling during connection authentication.
May 15 2015 : v1.5.1
Major maintenance release.
NOTICE: All LDTs on server other than LLIST have been deprecated, and will be removed in the future. As Such, all API regarding those features are considered deprecated and will be removed in tandem.
-
Improvements
-
Introduces
ClientPolicy.IdleTimeout
to close stale connections to the server. Thanks to Mário Freitas (@imkira). PR #57 -
Use type alias instead of struct for NullValue.
-
Removed workaround regarding filtering bin names on the client for
BatchGet
. Issue #60
-
-
Fixes
-
Fixed a few race conditions.
-
Fixed #58 regarding race condition accessing
Cluster.password
. -
Fixed minor bugs regarding handling of nulls in structs for
GetObj()
andPutObj()
. -
Fixed a bug regarding setting TaskIds on the client.
-
-
** Other Changes **
-
Removed deprecated
ReplaceRoles()
method. -
Removed deprecated
SetCapacity()
andGetCapacity()
methods for LDTs.
-
April 13 2015 : v1.5.0
This release includes potential BREAKING CHANGES.
-
New Features
- Introduces
ClientPolicy.LimitConnectionsToQueueSize
. If set to true, the client won't attemp to create new connections to the node if the total number of pooled connections to the node is equal or more than the pool size. The client will retry to poll a connection from the queue until a timeout occurs. If no timeout is set, it will only retry for ten times.
- Introduces
-
Improvements
-
BREAKING CHANGE: | Uses type aliases instead of structs in several XXXValue methods. This removes a memory allocation per
Value
usage. Since everyPut
operation uses at list one value object, this has the potential to improve application performance. Since the signature of severalNewXXXValue
methods have changed, this might break some existing code if you have used the value objects directly. -
Improved
Logger
so that it will accept a generalizedLogger
interface. Any Logger with aPrintf(format string, values ...interface{})
method can be used. Examples include Logrus. -
Improved
Client.BatchGet()
performance.
-
-
Fixes
-
Bin names were ignored in BatchCommands.
-
BatchCommandGet.parseRecord()
returned wrong values whenBinNames
was empty but not nil.
-
March 31 2015 : v1.4.2
Maintenance release.
-
Improvements
- Replace channel-based queue system with a lock-based algorithm.
- Marshaller now supports arrays of arbitrary types.
Client.GetObject()
now returns an error when the object is not found.- Partition calculation uses a trick that is twice as fast.
-
Improvements
- Unpacking BLOBs resulted in returning references to pooled buffers. Now copies are returned.
March 12 2015 : v1.4.1
This is a minor release to help improve the compatibility of the client on Mac OS, and to make cross compilation easier.
-
Improvements
- Node validator won't call net.HostLookup if an IP is passed as a seed to it.
Feb 17 2015 : v1.4.0
This is a major release, and makes using the client much easier to develop applications.
-
New Features
- Added Marshalling Support for Put and Get operations. Refer to Marshalling Test to see how to take advantage. Same functionality for other APIs will follow soon. Example:
type SomeStruct struct { A int `as:"a"` // alias the field to a Self *SomeStruct `as:"-"` // will not persist the field } type OtherStruct struct { i interface{} OtherObject *OtherStruct } obj := &OtherStruct { i: 15, OtherObject: OtherStruct {A: 18}, } key, _ := as.NewKey("ns", "set", value) err := client.PutObject(nil, key, obj) // handle error here rObj := &OtherStruct{} err = client.GetObject(nil, key, rObj)
- Added
Recordset.Results()
. Consumers of a recordset do not have to implement a select anymore. Instead of:
recordset, err := client.ScanAll(...) L: for { select { case r := <-recordset.Record: if r == nil { break L } // process record here case e := <-recordset.Errors: // handle error here } }
one should only range on
recordset.Results()
:recordset, err := client.ScanAll(...) for res := range recordset.Results() { if res.Err != nil { // handle error here } else { // process record here fmt.Println(res.Record.Bins) } }
Use of the old pattern is discouraged and deprecated, and direct access to recordset.Records and recordset.Errors will be removed in a future release.
-
Improvements
- Custom Types are now allowed as bin values.
Jan 26 2015 : v1.3.1
-
Improvements
- Removed dependency on
unsafe
package.
- Removed dependency on
Jan 20 2015 : v1.3.0
-
Breaking Changes
- Removed
Record.Duplicates
andGenerationPolicy/DUPLICATE
- Removed
-
New Features
-
Added Security Features: Please consult Security Docs on Aerospike website.
ClientPolicy.User
,ClientPolicy.Password
Client.CreateUser()
,Client.DropUser()
,Client.ChangePassword()
Client.GrantRoles()
,Client.RevokeRoles()
,Client.ReplaceRoles()
Client.QueryUser()
,Client.QueryUsers
-
Added
Client.QueryNode()
-
Added
ClientPolicy.TendInterval
-
-
Improvements
- Cleaned up Scan/Query/Recordset concurrent code
-
Fixes
-
Fixed a bug in
tools/cli/cli.go
. -
Fixed a bug when
GetHeaderOp()
would always translate intoGetOp()
-
Dec 29 2014: v1.2.0
-
New Features
- Added
NewKeyWithDigest()
method. You can now create keys with custom digests, or only using digests without knowing the original value. (Useful when you are getting back results with Query and Scan)
- Added
Dec 22 2014
-
New Features
-
Added
ConsistencyLevel
toBasePolicy
. -
Added
CommitLevel
toWritePolicy
. -
Added
LargeList.Range
andLargeList.RangeThenFilter
methods. -
Added
LargeMap.Exists
method.
-
-
Improvements
- We use a pooled XORShift RNG to produce random numbers in the client. It is FAST.
Dec 19 2014
-
Fixes
Record.Expiration
wasn't converted to TTL values onClient.BatchGet
,Client.Scan
andClient.Query
.
Dec 10 2014
-
Fixes:
- Fixed issue when the size of key field would not be estimated correctly when WritePolicy.SendKey was set.
Nov 27 2014
Major Performance Enhancements. Minor new features and fixes.
-
Improvements
-
Go client is much faster and more memory efficient now. In some workloads, it competes and wins against C and Java clients.
-
Complex objects are now de/serialized much faster.
-
-
New Features
- Added Default Policies for Client object. Instead of creating a new policy when the passed policy is nil, default policies will be used.
Nov 24 2014
-
Fixes:
- Fixed issue when WritePolicy.SendKey = true was not respected in Touch() and Operate()
Nov 22 2014
Hotfix in unpacker. Update strongly recommended for everyone using Complex objects, LDTs and UDFs.
-
Fixes:
- When Blob, ByteArray or String size has a bit sign set, unpacker reads it wrong. Note: This bug only affects unpacking of these objects. Packing was unaffected, and data in the database is valid.
Nov 2 2014
Minor, but very impoortant fix.
-
Fixes:
- Node selection in partition map was flawed on first refresh.
-
Incompatible changes:
Expiration
andGeneration
inWritePolicy
are nowint32
TaskId
inStatement
is now always set in the client, and isint64
-
New Features:
- float32, float64 and bool are now supported in map and array types
Oct 15 2014 (Beta 2)
-
Hot fix:
- Fixed pack/unpack for uint64
Aug 20 2014 (Beta 1)
Major changes and improvements.
-
New Features:
- Added client.Query()
- Added client.ScanNode()/All()
- Added client.Operate()
- Added client.CreateIndex()
- Added client.DropIndex()
- Added client.RegisterUDF()
- Added client.RegisterUDFFromFile()
- Added client.Execute()
- Added client.ExecuteUDF()
- Added client.BatchGet()
- Added client.BatchGetHeader()
- Added client.BatchExists()
- Added LDT implementation
- Added
Node
andKey
references to the Record
-
Changes:
-
Many minor and major bug fixes
-
Potentially breaking change: Reduced Undocumented API surface
-
Fixed a few places where error results were not checked
-
Breaking Change: Convert Key.namespace & Key.setName from pointer to string; affects Key API
-
Renamed all
this
receivers to appropriate names -
Major performance improvements (~2X improvements in speed and memory consumption):
- better memory management for commands; won't allocate if capacity is big enough
- better hash management in key; avoids two redundant memory allocs
- use a buffer pool to reduce GC load
- fine-grained, customizable and deterministic buffer pool implementation for command
-
Optimizations for Key & Digest
- changed digest implementation, removed an allocation
- Added RIPEMD160 hash files from crypto to lib
- pool hash objects
-
Various Benchmark tool improvements
- now profileable using localhost:6060
- minor bug fixes
-
Jul 26 2014 (Alpha)
- Initial Release.