- Tarsgo is high performance RPC framework in Golang programing language using the tars protocol.
- Go has become popular for programming with the rise of containerization technology such as docker, k8s, and etcd.
- Go's goroutine concurrency mechanism means Go is very suitable for large-scale high-concurrency back-end server program development. The Go language has nearly C/C++ performance and near Python productivity.
- In Tencent, part of the existing C++ development team has gradually turned into Go developers. Tars, a widely used RPC framework, supports C++, Java, NodeJS, and PHP, and now Go. The combination with Go language has become a general trend. Therefore, in the voice of users, we launched Tarsgo, and we have applied to Tencent map application, YingYongbao application, Internet plus and other projects.
- Learn more about the whole Tars architecture and design at Introduction.
- Tars2go tool: tars file is automatically generated and converted into Go language, contains RPC server/client code;
- Serialization and deserialization of tars protocol in Go.
- Auto service discovery.
- TCP/UDP/Http server & Client.
- The support of local logging and remote logging.
- The support of statistical reporting, property statistics, and anomaly reporting.
- The support of set division.
- The support of protocol buffers. See more in pb2tarsgo.
- The support of filter.
- The support of Zipkin OpenTracing.
- For installing OSS and other basic servers, see the Installation document.
- Install go 1.13.x or above (for example go install path:
/usr/local/go), setGOROOT,GOPATH, for example: in Linux:
export GOROOT=/usr/local/go
export GOPATH=/root/gocode
export PATH=$PATH:$GOROOT/bin:$GOPATH/binIf you in china, you can set go proxy:
go env -w GOPROXY=https://goproxy.cn Please set go mod:
go env -w GO111MODULE=autocd $GOPATH
go get -u github.com/TarsCloud/TarsGo/tarsnow tarsgo downloaded to path:
$GOPATH/src/github.com/TarsCloud/TarsGo/After tarsgo src be downloaded, please install tars2go:
go install $GOPATH/src/github.com/TarsCloud/TarsGo/tars/tools/tars2go- For quickstart, see tars_go_quickstart_en.md
- For performance, see tars_go_performance.md
- Below is a full example illustrating how to use tarsgo to build a server.
Create a tars file, like hello.tars, under $GOPATH/src (for example, $GOPATH/src/TestApp/TestServer/hello.tars).
For more detail about tars protocol, see tars_protocol Tars protocol is a binary, IDL-based protocol similar to protocol buffers.
module TestApp
{
interface Hello
{
int test();
int testHello(string sReq, out string sRsp);
};
}; If not install tars2go, compile and install the tars2go tools.
go install $GOPATH/src/github.com/TarsCloud/TarsGo/tars/tools/tars2gotars2go --outdir=./vendor hello.tarspackage main
import (
"github.com/TarsCloud/TarsGo/tars"
"TestApp"
)
type HelloImp struct {
}
//implete the Test interface
func (imp *HelloImp) Test() (int32, error) {
return 0, nil
}
//implete the testHello interface
func (imp *HelloImp) TestHello(in string, out *string) (int32, error) {
*out = in
return 0, nil
}
func main() { //Init servant
imp := new(HelloImp) //New Imp
app := new(TestApp.Hello) //New init the A Tars
cfg := tars.GetServerConfig() //Get Config File Object
app.AddServant(imp, cfg.App+"."+cfg.Server+".HelloObj") //Register Servant
tars.Run()
}Illustration:
- HelloImp is the struct where you implement the Hello & Test interface, notice that Test & Hello must start with an upper letter to be exported, which is the only place distinguished from tars file definition.
- TestApp.Hello is generated by
tar2gotools, which could be found in./vendor/TestApp/Hello_IF.go, with a package named TestApp which is the same as module TestApp in the tars file. tars.GetServerConfig()is used to get server-side configuration.cfg.App+"."+cfg.Server+".HelloObj"is the object name bind to the Servant, which the client will use this name to access the server.
tars.GetServerConfig() return a server config, which is defined as below:
type serverConfig struct {
Node string
App string
Server string
LogPath string
LogSize string
LogLevel string
Version string
LocalIP string
BasePath string
DataPath string
config string
notify string
log string
netThread int
Adapters map[string]adapterConfig
Container string
Isdocker bool
Enableset bool
Setdivision string
}- Node: local tarsnode address, only if you use tars platform to deploy will use this parameter.
- APP: The application name.
- Server: The server name.
- LogPath: The directory to save logs.
- LogSize: The size when rotate logs.
- LogLevel: The rotate log level.
- Version: Tarsgo version.
- LocalIP: Local IP address.
- BasePath: Base path for the binary.
- DataPath: Path for store some cache files.
- config: The configuration center for getting configuration, like
tars.tarsconfig.ConfigObj. - notify: The notify center for report notify report, like
tars.tarsnotify.NotifyObj. - log: The remote log centre, like
tars.tarslog.LogObj. - netThread: Reserved for controlling the go routine that receives and sends packages.
- Adapters: The specified configuration for each adapter.
- Container: Reserved for later use, to store the container name.
- Isdocker: Reserved for later use, to specify if the server running inside a container.
- Enableset: True if used set division.
- Setdivision: To specify which set division, like
gray.sz.*.
A server-side configuration looks like:
<tars>
<application>
enableset=Y
setdivision=gray.sz.*
<server>
node=tars.tarsnode.ServerObj@tcp -h 10.120.129.226 -p 19386 -t 60000
app=TestApp
server=HelloServer
localip=10.120.129.226
local=tcp -h 127.0.0.1 -p 20001 -t 3000
basepath=/usr/local/app/tars/tarsnode/data/TestApp.HelloServer/bin/
datapath=/usr/local/app/tars/tarsnode/data/TestApp.HelloServer/data/
logpath=/usr/local/app/tars/app_log/
logsize=10M
config=tars.tarsconfig.ConfigObj
notify=tars.tarsnotify.NotifyObj
log=tars.tarslog.LogObj
#timeout for deactiving, ms.
deactivating-timeout=2000
logLevel=DEBUG
</server>
</application>
</tars>An adapter represents a bind IP port for a certain object. app.AddServant(imp, cfg.App+"."+cfg.Server+".HelloObj") in the server implement code example, fulfill the binding of the adapter configuration and implement for the HelloObj.
A full example for an adapter, see below:
<tars>
<application>
<server>
#each adapter configuration
<TestApp.HelloServer.HelloObjAdapter>
#allow Ip for white list.
allow
# ip and port to listen on
endpoint=tcp -h 10.120.129.226 -p 20001 -t 60000
#handlegroup
handlegroup=TestApp.HelloServer.HelloObjAdapter
#max connection
maxconns=200000
#portocol, only tars for now.
protocol=tars
#max capbility in handle queue.
queuecap=10000
#timeout in ms for the request in the queue.
queuetimeout=60000
#servant
servant=TestApp.HelloServer.HelloObj
#threads in handle server side implement code. goroutine for golang.
threads=5
</TestApp.HelloServer.HelloObjAdapter>
</server>
</application>
</tars>The command to start the server:
./HelloServer --config=config.confSee below for a full example of config.conf, we will explain the client-side configuration later.
<tars>
<application>
enableset=n
setdivision=NULL
<server>
node=tars.tarsnode.ServerObj@tcp -h 10.120.129.226 -p 19386 -t 60000
app=TestApp
server=HelloServer
localip=10.120.129.226
local=tcp -h 127.0.0.1 -p 20001 -t 3000
basepath=/usr/local/app/tars/tarsnode/data/TestApp.HelloServer/bin/
datapath=/usr/local/app/tars/tarsnode/data/TestApp.HelloServer/data/
logpath=/usr/local/app/tars/app_log/
logsize=10M
config=tars.tarsconfig.ConfigObj
notify=tars.tarsnotify.NotifyObj
log=tars.tarslog.LogObj
deactivating-timeout=2000
logLevel=DEBUG
<TestApp.HelloServer.HelloObjAdapter>
allow
endpoint=tcp -h 10.120.129.226 -p 20001 -t 60000
handlegroup=TestApp.HelloServer.HelloObjAdapter
maxconns=200000
protocol=tars
queuecap=10000
queuetimeout=60000
servant=TestApp.HelloServer.HelloObj
threads=5
</TestApp.HelloServer.HelloObjAdapter>
</server>
<client>
locator=tars.tarsregistry.QueryObj@tcp -h 10.120.129.226 -p 17890
sync-invoke-timeout=3000
async-invoke-timeout=5000
refresh-endpoint-interval=60000
report-interval=60000
sample-rate=100000
max-sample-count=50
asyncthread=3
modulename=TestApp.HelloServer
</client>
</application>
</tars>Users can write a client-side code easily without writing any protocol-specified communicating code.
A client side example:
package main
import (
"fmt"
"github.com/TarsCloud/TarsGo/tars"
"TestApp"
)
//tars.Communicator should only init once and be global
var comm *tars.Communicator
func main() {
comm = tars.NewCommunicator()
obj := "TestApp.TestServer.HelloObj@tcp -h 127.0.0.1 -p 10015 -t 60000"
app := new(TestApp.Hello)
comm.StringToProxy(obj, app)
var req string="Hello World"
var out string
ret, err := app.TestHello(req, &out)
if err != nil {
fmt.Println(err)
return
}
fmt.Println(ret, out)
}Illustration:
- package TestApp was generated by
tars2gotool using the tars protocol file. - comm: A Communicator is used for communicating with the server side which should only init once and be global.
- obj: object name which to specify the server IP and port. Usually, we just need the object name before the "@" character.
- app: Application that associated with the interface in the tars file. In this case, it's
TestApp.Hello. StringToProxy:StringToProxymethod is used for binding the object name and the application, if skip this, the communicator won't know who to communicate with for the application.req, res: In and out parameter which define in the tars file for the TestHello method.app.TestHellois used to call the method defined in the tars file, and ret and err will be returned.
A communicator represents a group of resources for sending and receiving packages for the client-side, which in the end manages the socket communicating for each object.
You will only need one communicator in a program.
var comm *tars.Communicato
comm = tars.NewCommunicator()
comm.SetProperty("property", "tars.tarsproperty.PropertyObj")
comm.SetProperty("locator", "tars.tarsregistry.QueryObj@tcp -h ... -p ...")Description:
- The communicator's configuration file format will be described later.
- Communicators can be configured without a configuration file, and all parameters have default values.
- The communicator can also be initialized directly through the
SetPropertymethod.- If you don't want a configuration file, you must set the locator parameter yourself.
Communicator attribute description:
- locator: The address of the registry service must be in the format
ip port. If you do not need the registry to locate the service, you do not need to configure this item.- important async-invoke-timeout: The maximum timeout (in milliseconds) for client calls. The default value for this configuration is
3000.- sync-invoke-timeout: Unused for tarsgo right now.
- refresh-endpoint-interval: The interval (in milliseconds) for periodically accessing the registry to obtain information. The default value for this configuration is one minute.
- stat: The address of the service is called between modules. If this item is not configured, it means that the reported data will be directly discarded.
- property: The address that the service reports its attribute. If it is not configured, this means that the reported data is directly discarded.
- report-interval: Unused for tarsgo for now.
- asyncthread: Discarded for tarsgo.
- modulename: The module name, the default value is the name of the executable program.
The format of the communicator's configuration file is as follows:
<tars>
<application>
#The configuration required by the proxy
<client>
#address
locator = tars.tarsregistry.QueryObj@tcp -h 127.0.0.1 -p 17890
#The maximum timeout (in milliseconds) for synchronous calls.
sync-invoke-timeout = 3000
#The maximum timeout (in milliseconds) for asynchronous calls.
async-invoke-timeout = 5000
#The maximum timeout (in milliseconds) for synchronous calls.
refresh-endpoint-interval = 60000
#Used for inter-module calls
stat = tars.tarsstat.StatObj
#Address used for attribute reporting
property = tars.tarsproperty.PropertyObj
#report time interval
report-interval = 60000
#The number of threads that process asynchronous responses
asyncthread = 3
#The module name
modulename = Test.HelloServer
</client>
</application>
</tars>if you want to use timeout control on the client-side, use TarsSetTimeout which in ms.
app := new(TestApp.Hello)
comm.StringToProxy(obj, app)
app.TarsSetTimeout(3000)This section details how the Tars client remotely invokes the server.
First, briefly describe the addressing mode of the Tars client. Secondly, it will introduce the calling method of the client, including but not limited to one-way calling, synchronous calling, asynchronous calling, hash calling, and so on.
The addressing mode of the Tars service can usually be divided into two ways: the service name is registered in the master and the service name is not registered in the master. A master is a name server (routing server) dedicated to registering service node information.
The service name added in the name server is implemented through the operation management platform.
For services that are not registered with the master, it can be classified as direct addressing, that is, the IP address of the service provider needs to be specified before calling the service. The client needs to specify the specific address of the HelloObj object when calling the service:
that is Test.HelloServer.HelloObj@tcp -h 127.0.0.1 -p 9985
Test.HelloServer.HelloObj: Object name
tcp: Tcp protocol
-h: Specify the host address, here is 127.0.0.1
-p: Port, here is 9985
If HelloServer is running on two servers, the app is initialized as follows:
obj:= "Test.HelloServer.HelloObj@tcp -h 127.0.0.1 -p 9985:tcp -h 192.168.1.1 -p 9983"
app := new(TestApp.Hello)
comm.StringToProxy(obj, app)The address of HelloObj is set to the address of the two servers. At this point, the request will be distributed to two servers (distribution method can be specified, not introduced here). If one server is down, the request will be automatically assigned to another one, and the server will be restarted periodically.
For services registered in the master, the service is addressed based on the service name. When the client requests the service, it does not need to specify the specific address of the HelloServer, but it needs to specify the address of the registry when generating the communicator or initializing the communicator.
The following shows the address of the registry by setting the parameters of the communicator:
var *tars.Communicator
comm = tars.NewCommunicator()
comm.SetProperty("locator", "tars.tarsregistry.QueryObj@tcp -h ... -p ...")Since the client needs to rely on the registry's address, the registry must also be fault-tolerant. The registry's fault-tolerant method is the same as above, specifying the address of the two registries.
TODO. Unsupported yet in tarsgo.
package main
import (
"fmt"
"github.com/TarsCloud/TarsGo/tars"
"TestApp"
)
var *tars.Communicator
func main() {
comm = tars.NewCommunicator()
obj := "TestApp.TestServer.HelloObj@tcp -h 127.0.0.1 -p 10015 -t 60000"
app := new(TestApp.Hello)
comm.StringToProxy(obj, app)
var req string="Hello World"
var out string
ret, err := app.TestHello(req, &out)
if err != nil {
fmt.Println(err)
return
}
fmt.Println(ret, out)
}tarsgo can use Asynchronous calls easily using goroutine. Unlike C++, we don't need to implement a callback function.
package main
import (
"fmt"
"github.com/TarsCloud/TarsGo/tars"
"time"
"TestApp"
)
var *tars.Communicator
func main() {
comm = tars.NewCommunicator()
obj := "TestApp.TestServer.HelloObj@tcp -h 127.0.0.1 -p 10015 -t 60000"
app := new(TestApp.Hello)
comm.StringToProxy(obj, app)
go func(){
var req string="Hello World"
var out string
ret, err := app.TestHello(req, &out)
if err != nil {
fmt.Println(err)
return
}
fmt.Println(ret, out)
}()
time.Sleep(1)
}The client can call server by set through configuration file mentioned. Which enableset will be y and setdivision will set like gray.sz.*. See https://github.com/TarsCloud/Tars/blob/master/docs-en/tars_idc_set.md for more detail.
If you want to call by set manually, tarsgo will support this feature soon.
Since multiple servers can be deployed, client requests are randomly distributed to the server, but in some cases, certain requests should be always sent to a particular server. In this case, Tars provides a simple way to achieve which is called hash-call. Tarsgo will support this feature soon.
//Define the return code given by the TARS service
const int TARSSERVERSUCCESS = 0; //Server-side processing succeeded
const int TARSSERVERDECODEERR = -1; //Server-side decoding exception
const int TARSSERVERENCODEERR = -2; //Server-side encoding exception
const int TARSSERVERNOFUNCERR = -3; //There is no such function on the server side
const int TARSSERVERNOSERVANTERR = -4; //The server does not have the Servant object
const int TARSSERVERRESETGRID = -5; // server grayscale state is inconsistent
const int TARSSERVERQUEUETIMEOUT = -6; //server queue exceeds limit
const int TARSASYNCCALLTIMEOUT = -7; // Asynchronous call timeout
const int TARSINVOKETIMEOUT = -7; //call timeout
const int TARSPROXYCONNECTERR = -8; //proxy link exception
const int TARSSERVEROVERLOAD = -9; //Server overload, exceeding queue length
const int TARSADAPTERNULL = -10; //The client routing is empty, the service does not exist or all services are down.
const int TARSINVOKEBYINVALIDESET = -11; //The client calls the set rule illegally
const int TARSCLIENTDECODEERR = -12; //Client decoding exception
const int TARSSERVERUNKNOWNERR = -99; //The server is in an abnormal positionA quick example for using tarsgo rotating log:
TLOG := tars.GetLogger("TLOG")
TLOG.Debug("Debug logging")This is will create a *Rogger.Logger, which was defined in tars/util/rogger, and after GetLogger was called, and a logfile is created under Logpath defined in the config.conf which name is cfg.App + "." + cfg.Server + "_" + name, and will be rotated after 100MB(default), and max rotated file is 10(default).
If you don't want to rotate log by file size. For example, you want to rotate by day, then use:
TLOG := tars.GetDayLogger("TLOG",1)
TLOG.Debug("Debug logging")For rotating by hour, use GetHourLogger("TLOG",1).
If you want to log to a remote server, which is defined in config.conf named tars.tarslog.LogObj. A full tars file definition can be found in tars/protocol/res/LogF.tars. You have to setup a log server before doing this. A log server can be found under Tencent/Tars/cpp/framework/LogServer. A quick example:
TLOG := GetRemoteLogger("TLOG")
TLOG.Debug("Debug logging")If you want to set the log level, you can set it from the OSS platform provided by tars project under Tencent/Tars/web.
If you want to customize your logger, see more detail in tars/util/rogger, tars/logger.go and tars/remotelogger.go.
The Tars server framework supports dynamic receiving commands to handle related business logic, such as dynamic update configuration.
tarsgo currently has tars.viewversion / tars.setloglevel administration commands for now. Users can send admin command from OSS to see what version is or setting log level mentioned about.
If you want to defined ur own admin commands, see this example:
func helloAdmin(who string ) (string, error) {
return who, nil
}
tars.RegisterAdmin("tars.helloAdmin", helloAdmin)Then you can send self-defined admin command tars.helloAdmin tarsgo and tarsgo will be shown in the browser.
Illustration:
// A function should be in this format
type adminFn func(string) (string, error)
//then u should registry this function using
func RegisterAdmin(name string, fn adminFn)Reporting statistics information is the logic of reporting the time-consuming information and other information to tarsstat inside the Tars framework. No user development is required. After the relevant information is correctly set during program initialization, it can be automatically reported inside the framework (including the client and the server).
After the client called the reporting interface, it is temporarily stored in memory. When it reaches a certain time point, it is reported to the tarsstat service (the default is once reporting 1 minute). We call the time gap between the two reporting time points as a statistical interval and perform the operations such as accumulating and comparing the same key in a statistical interval.
The sample code is as follows:
//for error
ReportStat(msg, 0, 1, 0)
//for success
ReportStat(msg, 1, 0, 0)
//func ReportStat(msg *Message, succ int32, timeout int32, exec int32)
//see more detail in tars/statf.goDescription:
- Normally, we don't have to concern about the Statistical reporting, the tarsgo framework will do this report after every client call server, no matter success or failure. And the success rate, fail rate, average cost time, and so on will be shown in the web management system if you set up properly.
- If the main service is deployed on the web management system, you do not need to define Communicator set the configurations of
tarsregistry,tarsstat, etc., the service will be automatically reported.- If the main service or program is not deployed on the web management system, you need to define the Communicator, set the
tarsregistry,tarsstat, etc., so that you can view the service monitoring of the called service on the web management system.- The reported data is reported regularly and can be set in the configuration of the communicator.
For better monitoring, the TARS framework supports reporting abnormal situation directly to tarsnotify in the program and can be viewed on the Web management page.
The framework provides three macros to report different kinds of exceptions:
tars.reportNotifyInfo("Get data from mysql error!")Info is a string, which can directly report the string to tarsnotify. The reported string can be seen on the page, subsequently, we can alarm according to the reported information.
To facilitate business statistics, the TARS framework also supports the display of information on the web management platform.
The types of statistics currently supported include the following:
Sum(sum): calculate the sum of every report value.Average(avg): calculate the average of every report value.Distribution(distr): calculate the distribution of every report, which parameter is a list, and calculate the probability distribution of each interval.Maximum(max): calculate the maximum of every report value.Minimum(min): calculate the minimum of every report value.Count(count): calculate the count of report times.
The sample code is the following:
sum := tars.NewSum()
count := tars.NewCount()
max := tars.NewMax()
min := tars.NewMin()
d := []int{10, 20, 30, 50}
distr := tars.NewDistr(d)
p := tars.CreatePropertyReport("testproperty", sum, count, max, min, distr)
for i := 0; i < 5; i++ {
v := rand.Intn(100)
p.Report(v)
}Description:
- Data is reported regularly, and can be set in the configuration of the communicator, currently once per minute.
- Create a
PropertyReportPtrfunction: The parametercreatePropertyReportcan be any collection of statistical methods, the example uses six statistical methods, usually only need to use one or two.- Note that when you call
createPropertyReport, you must create and save the created object after the service is enabled, and then just take the object to report, do not create it each time you use.
Users can set up remote configuration from OSS. See more detail in https://github.com/TarsCloud/TarsFramework/blob/master/docs-en/tars_config.md.
That is an example to illustrate how to use this API to get a configuration file from remote.
import "github.com/TarsCloud/TarsGo/tars"
...
cfg := tars.GetServerConfig()
remoteConf := tars.NewRConf(cfg.App, cfg.Server, cfg.BasePath)
config, _ := remoteConf.GetConfig("test.conf")
...setting.go in package tars is used to control tarsgo performance and characteristics. Some options should be updated from Getserverconfig().
//number of worker routines to handle client request
//zero means no control, just one goroutine for a client request.
//runtime.NumCpu() usually best performance in the benchmark.
var MaxInvoke int = 0
const (
//for now, some option should update from remote config
//version
TarsVersion string = "1.0.0"
//server
AcceptTimeout time.Duration = 500 * time.Millisecond
//zero for not set read deadline for Conn (better performance)
ReadTimeout time.Duration = 0 * time.Millisecond
//zero for not set write deadline for Conn (better performance)
WriteTimeout time.Duration = 0 * time.Millisecond
//zero for not set deadline for invoke user interface (better performance)
HandleTimeout time.Duration = 0 * time.Millisecond
IdleTimeout time.Duration = 600000 * time.Millisecond
ZombileTimeout time.Duration = time.Second * 10
QueueCap int = 10000000
//client
ClientQueueLen int = 10000
ClientIdleTimeout time.Duration = time.Second * 600
ClientReadTimeout time.Duration = time.Millisecond * 100
ClientWriteTimeout time.Duration = time.Millisecond * 3000
ReqDefaultTimeout int32 = 3000
ObjQueueMax int32 = 10000
//report
PropertyReportInterval time.Duration = 10 * time.Second
StatReportInterval time.Duration = 10 * time.Second
//mainloop
MainLoopTicker time.Duration = 10 * time.Second
//adapter
AdapterProxyTicker time.Duration = 10 * time.Second
AdapterProxyResetCount int = 5
//communicator default, update from remote config
refreshEndpointInterval int = 60000
reportInterval int = 10000
AsyncInvokeTimeout int = 3000
//tcp network config
TCPReadBuffer = 128 * 1024 * 1024
TCPWriteBuffer = 128 * 1024 * 1024
TCPNoDelay = false
)tars.TarsHttpMux is multiplexer like http.ServeMux,the pattern parameter is used as the interface name in monitoring report.
Here is a sample of HTTP server:
package main
import (
"net/http"
"github.com/TarsCloud/TarsGo/tars"
)
func main() {
mux := &tars.TarsHttpMux{}
mux.HandleFunc("/", func(w http.ResponseWriter, r *http.Request) {
w.Write([]byte("Hello tafgo"))
})
cfg := tars.GetServerConfig()
tars.AddHttpServant(mux, cfg.App+"."+cfg.Server+".HttpObj") //Register http server
tars.Run()
}In the past, TarsGo did not use context in the generated client code, or the implementation code passed in by the user. This makes us want to pass some framework information, such as client IP, port, etc., or the user passes some information about the call chain to the framework, which is difficult to implement. Through a refactoring of the interface, the context is supported, and these information will be implemented through the context. This refactoring is designed to be compatible with older user behavior and is fully compatible.
Server-Side Context:
type ContextTestImp struct {
}
//only need to add ctx context.Context parameter
func (imp *ContextTestImp) Add(ctx context.Context, a int32, b int32, c *int32) (int32, error) {
//We can use context to get some usefull infomation we need, such Client, ip, port and tracing infomation
//read more detail under tars/util/current
ip, ok := current.GetClientIPFromContext(ctx)
if !ok {
logger.Error("Error getting ip from context")
}
return 0, nil
}
//just change AddServant into AddServantWithContext
app.AddServantWithContext(imp, cfg.App+"."+cfg.Server+".ContextTestObj")Client-Side Context:
ctx := context.Background()
c := make(map[string]string)
c["a"] = "b"
//juse change app.Add into app.AddWithContext, now you can pass context to framework,
//if you want to setting request package's context, you can pass a optional parameter, just like c, which is ...[string]string
ret, err := app.AddWithContext(ctx, i, i*2, &out, c)Read full demo client and server under _examples/ContextTestServer
For supporting the writing plugin, we provide the filter concept to the framework. We have the client-side filter and the server-side filter.
//ServerFilter, dispatch and f is passed as parameter, for dispatching user's implement.
//req and resp is
type ServerFilter func(ctx context.Context, d Dispatch, f interface{}, req *requestf.RequestPacket, resp *requestf.ResponsePacket, withContext bool) (err error)
//
type ClientFilter func(ctx context.Context, msg *Message, invoke Invoke, timeout time.Duration) (err error)
//RegisterServerFilter registers the server side filter
//func RegisterServerFilter(f ServerFilter)
//RegisterClientFilter registers the client side filter
//func RegisterClientFilter(f ClientFilter)Having these filters, now we can add OpenTracing for every request. Let's take a look at the client-side filter for OpenTracing.
//ZipkinClientFilter returns a client side tars filter, for hooking zipking opentracing.
func ZipkinClientFilter() tars.ClientFilter {
return func(ctx context.Context, msg *tars.Message, invoke tars.Invoke, timeout time.Duration) (err error) {
var pCtx opentracing.SpanContext
req := msg.Req
//If span context is passed in the context, we use this context as parent span, else start a new span.
//The method name of the rpc request, is used as span's name.
if parent := opentracing.SpanFromContext(ctx); parent != nil {
pCtx = parent.Context()
}
cSpan := opentracing.GlobalTracer().StartSpan(
req.SFuncName,
opentracing.ChildOf(pCtx),
ext.SpanKindRPCClient,
)
defer cSpan.Finish()
cfg := tars.GetServerConfig()
//set additional information for the span, like method, interface, protocol, vesion, ip and port etc.
cSpan.SetTag("client.ipv4", cfg.LocalIP)
cSpan.SetTag("tars.interface", req.SServantName)
cSpan.SetTag("tars.method", req.SFuncName)
cSpan.SetTag("tars.protocol", "tars")
cSpan.SetTag("tars.client.version", tars.TarsVersion)
//inject the span context into the request package's status, which is map[string]string
if req.Status != nil {
err = opentracing.GlobalTracer().Inject(cSpan.Context(), opentracing.TextMap, opentracing.TextMapCarrier(req.Status))
if err != nil {
logger.Error("inject span to status error:", err)
}
} else {
s := make(map[string]string)
err = opentracing.GlobalTracer().Inject(cSpan.Context(), opentracing.TextMap, opentracing.TextMapCarrier(s))
if err != nil {
logger.Error("inject span to status error:", err)
} else {
req.Status = s
}
}
//Nothing tho change, just invoke the request.
err = invoke(ctx, msg, timeout)
if err != nil {
//invoke error, logging the error information to the span.
ext.Error.Set(cSpan, true)
cSpan.LogFields(oplog.String("event", "error"), oplog.String("message", err.Error()))
}
return err
}The server will add filters, which exact the span context from the request package's status and start a new span.
Read more under TarsGo/tars/plugin/zipkintracing. For client-side and server-side example code, read ZipkinTraceClient & ZipkinTraceServer under the examples.