The etcd-cpp-apiv3 is a C++ API for etcd's v3 client API,
i.e., ETCDCTL_API=3
.
- Linux
- Ubuntu 18.04, requires upgrade grpc libraries (has been tested with 1.27.x).
- Ubuntu 20.04
- MacOS
- MacOS 10.15
- MacOS 11.0
- Windows
- Windows 10, with vcpkg
-
boost
-
On Ubuntu, above requirement could be installed as:
apt-get install libboost-all-dev
-
On MacOS, above requirement could be installed as:
brew install boost
-
-
protobuf
-
gRPC
-
On Ubuntu, above requirements related to protobuf and gRPC can be installed as:
apt-get install libgrpc-dev \ libgrpc++-dev \ libprotobuf-dev \ protobuf-compiler-grpc
-
On MacOS, above requirements related to protobuf and gRPC can be installed as:
brew install grpc protobuf
-
-
cpprestsdk, the latest version of master branch on github should work, you can build and install this dependency using cmake with:
git clone https://github.com/microsoft/cpprestsdk.git cd cpprestsdk mkdir build && cd build cmake .. -DCPPREST_EXCLUDE_WEBSOCKETS=ON make -j && make install
The etcd-cpp-apiv3 library could be easily built and installed using cmake, after all above dependencies have been successfully installed:
git clone https://github.com/etcd-cpp-apiv3/etcd-cpp-apiv3.git
cd etcd-cpp-apiv3
mkdir build && cd build
cmake ..
make -j && make install
The etcd-cpp-apiv3 should work well with etcd > 3.0. Feel free to issue an issue to us on Github when you encounter problems when working with etcd 3.x releases.
etcd::Client etcd("http://127.0.0.1:4001");
etcd::Response response = etcd.get("/test/key1").get();
std::cout << response.value().as_string();
Methods of the etcd client object are sending the corresponding gRPC requests and are returning
immediatelly with a pplx::task
object. The task object is responsible for handling the
reception of the HTTP response as well as parsing the gRPC of the response. All of this is done
asynchronously in a background thread so you can continue your code to do other operations while the
current etcd operation is executing in the background or you can wait for the response with the
wait()
or get()
methods if a synchron behaviour is enough for your needs. These methods
are blocking until the HTTP response arrives or some error situation happens. get()
method
also returns the etcd::Response
object.
etcd::Client etcd("http://127.0.0.1:4001");
pplx::task<etcd::Response> response_task = etcd.get("/test/key1");
// ... do something else
etcd::Response response = response_task.get();
std::cout << response.value().as_string();
The pplx library allows to do even more. You can attach continuation ojects to the task if you do
not care about when the response is coming you only want to specify what to do then. This
can be achieved by calling the then
method of the task, giving a funcion object parameter to
it that can be used as a callback when the response is arrived and processed. The parameter of this
callback should be either a etcd::Response
or a pplx::task<etcd:Response>
. You should
probably use a C++ lambda funcion here as a callback.
etcd::Client etcd("http://127.0.0.1:4001");
etcd.get("/test/key1").then([](etcd::Response response)
{
std::cout << response.value().as_string();
});
// ... your code can continue here without any delay
Your lambda function should have a parameter of type etcd::Response
or
pplx::task<etcd::Response>
. In the latter case you can get the actual etcd::Response
object with the get()
function of the task. Calling get can raise exeptions so this is the way
how you can catch the errors generated by the REST interface. The get()
call will not block in
this case since the respose has been already arrived (we are inside the callback).
etcd::Client etcd("http://127.0.0.1:4001");
etcd.get("/test/key1").then([](pplx::task<etcd::Response> response_task)
{
try
{
etcd::Response response = response.task.get(); // can throw
std::cout << response.value().as_string();
}
catch (std::ecxeption const & ex)
{
std::cerr << ex.what();
}
});
// ... your code can continue here without any delay
Connecting to multiple endpoints is supported:
// multiple endpoints are separated by comma
etcd::Client etcd("http://a.com:2379,http://b.com:2379,http://c.com:2379");
// or, separated semicolon
etcd::Client etcd("http://a.com:2379;http://b.com:2379;http://c.com:2379");
Behind the screen, gRPC's load balancer is used and the round-robin strategy will be used by default.
Etcd v3 authentication has been
supported. The Client::Client
could accept a username
and password
as arguments and handle
the authentication properly.
etcd::Client etcd("http://127.0.0.1:2379", "root", "root");
Or the etcd client can be constructed explictly:
etcd::Client *etcd = etcd::Client::WithUser(
"http://127.0.0.1:2379", "root", "root");
Enabling v3 authentication requires a bit more work for older versions etcd (etcd 3.2.x and etcd 3.3.x).
First you need to set the ETCDCTL_API=3
, then
- add a user, and type the password:
printf 'root\nroot\n' | /usr/local/bin/etcdctl user add root
- enabling authentication:
/usr/local/bin/etcdctl auth enable
- disable authentication:
/usr/local/bin/etcdctl --user="root:root" auth disable
Etcd transport security and certificate based
authentication have been supported as well. The Client::Client
could accept arguments ca
,
cert
and key
for CA cert, cert and private key files for the SSL/TLS transport and authentication.
Note that the later arguments cert
and key
could be empty strings or omitted if you just need
secure transport and don't enable certificate-based client authentication (using the --client-cert-auth
arguments when launching etcd server).
etcd::Client etcd("https://127.0.0.1:2379",
"example.rootca.cert", "example.cert", "example.key",
"round_robin");
Or the etcd client can be constructed explictly:
etcd::Client *etcd = etcd::Client::WithSSL(
"https://127.0.0.1:2379",
"example.rootca.cert", "example.cert", "example.key");
Using secure transport but not certificated-based client authentication:
etcd::Client *etcd = etcd::Client::WithSSL(
"https://127.0.0.1:2379", "example.rootca.cert");
For more details about setup about security communication between etcd server and client, please refer to transport security in etcd documentation and an example about setup etcd with transport security using openssl.
We also provide a tool setup-ca.sh
as a helper for development and testing.
You can read a value with the get
method of the clinent instance. The only parameter is the
key to be read. If the read operation is successful then the value of the key can be acquired with
the value()
method of the response. Success of the operation can be checked with the
is_ok()
method of the response. In case of an error, the error_code()
and
error_message()
methods can be called for some further detail.
Please note that there can be two kind of error situations. There can be some problem with the
communication between the client and the etcd server. In this case the get()
method of the
response task will throw an exception as shown above. If the communication is ok but there is some
problem with the content of the actual operation, like attemp to read a non-existing key then the
response object will give you all the details. Let's see this in an example.
The Value object of the response also holds some extra information besides the string value of the
key. You can also get the index number of the creation and the last modification of this key with
the created_index()
and the modofied_index()
methods.
etcd::Client etcd("http://127.0.0.1:4001");
pplx::task<etcd::Response> response_task = etcd.get("/test/key1");
try
{
etcd::Response response = response_task.get(); // can throw
if (response.is_ok())
std::cout << "successful read, value=" << response.value().as_string();
else
std::cout << "operation failed, details: " << response.error_message();
}
catch (std::ecxeption const & ex)
{
std::cerr << "communication problem, details: " << ex.what();
}
Setting the value of a key can be done with the set()
method of the client. You simply pass
the key and the value as string parameters and you are done. The newly set value object can be asked
from the response object exactly the same way as in case of the reading (with the value()
method). This way you can check for example the index value of your modification. You can also check
what was the previous value that this operation was overwritten. You can do that with the
prev_value()
method of the response object.
etcd::Client etcd("http://127.0.0.1:4001");
pplx::task<etcd::Response> response_task = etcd.set("/test/key1", "42");
try
{
etcd::Response response = response_task.get();
if (response.is_ok())
std::cout << "The new value is successfully set, previous value was "
<< response.prev_value().as_string();
else
std::cout << "operation failed, details: " << response.error_message();
}
catch (std::ecxeption const & ex)
{
std::cerr << "communication problem, details: " << ex.what();
}
The set method creates a new leaf node if it weren't exists already or modifies an existing one. There are a couple of other modification methods that are executing the write operation only upon some specific conditions.
add(key, value)
creates a new value if it's key does not exists and returns a "Key already exists" error otherwise (error code 105)modify(key, value)
modifies an already existing value or returns a "Key not found" error otherwise (error code 100)modify_if(key, value, old_value)
modifies an already existing value but only if the previous value equals with old_value. If the values does not match returns with "Compare failed" error (code 101)modify_if(key, value, old_index)
modifies an already existing value but only if the index of the previous value equals with old_index. If the indices does not match returns with "Compare failed" error (code 101)
Values can be deleted with the rm
method passing the key to be deleted as a parameter. The key
should point to an existing value. There are conditional variations for deletion too.
rm_if(key, value, old_value)
deletes an already existing value but only if the previous value equals with old_value. If the values does not match returns with "Compare failed" error (code 101)rm_if(key, value, old_index)
deletes an already existing value but only if the index of the previous value equals with old_index. If the indices does not match returns with "Compare failed" error (code 101)
Directory nodes are not supported anymore in etcdv3. However, ls and rmdir will list/delete keys defined by the prefix. mkdir method is removed since etcdv3 treats everything as keys.
-
Creating a directory:
Creating a directory is not supported anymore in etcdv3 cpp client. Users should remove the API from their code.
-
Listing a directory:
Listing directory in etcd3 cpp client will return all keys that matched the given prefix recursively.
etcd.set("/test/key1", "value1").wait();
etcd.set("/test/key2", "value2").wait();
etcd.set("/test/key3", "value3").wait();
etcd.set("/test/subdir/foo", "foo").wait();
etcd::Response resp = etcd.ls("/test/new_dir").get();
resp.key()
will have the following values:
/test/key1
/test/key2
/test/key3
/test/subdir/foo
Note: Regarding the returned keys when listing a directory:
- In etcdv3 cpp client, resp.key(0) will return "/test/new_dir/key1" since everything is treated as keys in etcdv3.
- While in etcdv2 cpp client it will return "key1" and "/test/new_dir" directory should be created first before you can set "key1".
When you list a directory the response object's keys()
and values()
methods gives
you a vector of key names and values. The value()
method with an integer parameter also
returns with the i-th element of the values vector, so response.values()[i] == response.value(i)
.
etcd::Client etcd("http://127.0.0.1:4001");
etcd::Response resp = etcd.ls("/test/new_dir").get();
for (int i = 0; i < resp.keys().size(); ++i)
{
std::cout << resp.keys(i);
std::cout << " = " << resp.value(i).as_string() << std::endl;
}
- Removing directory:
If you want the delete recursively then you have to pass a second true
parameter
to rmdir and supply a key. This key will be treated as a prefix. All keys that match the
prefix will be deleted. All deleted keys will be placed in response.values()
and
response.keys()
. This parameter defaults to false
.
etcd::Client etcd("http://127.0.0.1:4001");
etcd.set("/test/key1", "foo");
etcd.set("/test/key2", "bar");
etcd.set("/test/key3", "foo_bar");
etcd::Response resp = etcd.rmdir("/test", true).get();
for (int i = 0; i < resp.keys().size(); ++i)
{
std::cout << resp.keys(i);
std::cout << " = " << resp.value(i).as_string() << std::endl;
}
However, if recursive parameter is false, functionality will be the same as just deleting a key. The key supplied will NOT be treated as a prefix and will be treated as a normal key name.
Etcd lock has been supported as follows:
etcd::Client etcd("http://127.0.0.1:4001");
etcd.lock("/test/lock");
It will create a lease and a keep-alive job behind the screen, the lease will be revoked until the lock is unlocked.
Users can also feed their own lease directory for lock:
etcd::Client etcd("http://127.0.0.1:4001");
etcd.lock_with_lease("/test/lock", lease_id);
Watching for a change is possible with the watch()
operation of the client. The watch method
simply does not deliver a response object until the watched value changes in any way (modified or
deleted). When a change happens the returned result object will be the same as the result object of
the modification operation. So if the change is triggered by a value change, then
response.action()
will return "set", response.value()
will hold the new
value and response.prev_value()
will contain the previous value. In case of a delete
response.action()
will return "delete", response.value()
will be empty and should not be
called at all and response.prev_value()
will contain the deleted value.
As mentioned in the section "handling directory nodes", directory nodes are not supported anymore in etcdv3.
However it is still possible to watch a whole "directory subtree", or more specifically a set of
keys that match the prefix, for changes with passing true
to the second recursive
parameter of watch
(this parameter defaults to false
if omitted). In this case the
modified value object's key()
method can be handy to determine what key is actually changed.
Since this can be a long lasting operation you have to be prepared that is terminated by an
exception and you have to restart the watch operation.
The watch also accepts an index parameter that specifies what is the first change we are interested about. Since etcd stores the last couple of modifications with this feature you can ensure that your client does not miss a single change.
Here is an example how you can watch continuously for changes of one specific key.
void watch_for_changes()
{
etcd.watch("/nodes", index + 1, true).then([this](pplx::task<etcd::Response> resp_task)
{
try
{
etcd::Response resp = resp_task.get();
index = resp.index();
std::cout << resp.action() << " " << resp.value().as_string() << std::endl;
}
catch(...) {}
watch_for_changes();
});
}
At first glance it seems that watch_for_changes()
calls itself on every value change but in
fact it just sends the asynchron request, sets up a callback for the response and then returns. The
callback is executed by some thread from the pplx library's thread pool and the callback (in this
case a small lambda function actually) will call watch_for_changes
again from there.
Users can watch a key indefinitely or until user cancels the watch. This can be done by
instantiating a Watcher class. The supplied callback function in Watcher class will be
called every time there is an event for the specified key. Watch stream will be cancelled
either by user implicitly calling Cancel()
or when watcher class is destroyed.
etcd::Watcher watcher("http://127.0.0.1:2379", "/test", printResponse);
etcd.set("/test/key", "42"); /* print response will be called */
etcd.set("/test/key", "43"); /* print response will be called */
watcher.Cancel();
etcd.set("/test/key", "43"); /* print response will NOT be called,
since watch is already cancelled */
}
Users can request for lease which is governed by a time-to-live(TTL) value given by the user.
Moreover, user can attached the lease to a key(s) by indicating the lease id in add()
,
set()
, modify()
and modify_if()
. Also the ttl will that was granted by etcd
server will be indicated in ttl()
.
etcd::Client etcd("http://127.0.0.1:4001");
etcd::Response resp = etcd.leasegrant(60).get();
etcd.set("/test/key2", "bar", resp.value().lease());
std::cout << "ttl" << resp.value().ttl();
The lease can be revoked by
etcd.leaserevoke(resp.value().lease());
A lease can also be attached with a KeepAlive
object at the creation time,
std::shared_ptr<etcd::KeepAlive> keepalive = etcd.leasekeepalive(60).get();
std::cout << "lease id: " << keepalive->Lease();
The remaining time-to-live of a lease can be inspected by
etcd::Response resp2 = etcd.leasetimetolive(resp.value().lease()).get();
std::cout << "ttl" << resp.value().ttl();
Keep alive for leases is implemented using a seperate class KeepAlive
, which can be used as:
etcd::KeepAlive keepalive(etcd, ttl, lease_id);
It will perform a periodly keep-alive action before it is cancelled explicitly, or destructed implicitly.
KeepAlive
may fails (e.g., when the etcd server stopped unexpectedly), the constructor of KeepAlive
could accept a handler of type std::function<std::exception_ptr>
and the handler will be invoked
when exception occurs during keeping it alive.
Note that the handler will invoked in a separated thread, not the thread where the KeepAlive
object
is constructed.
std::function<void (std::exception_ptr)> handler = [](std::exception_ptr eptr) {
try {
if (eptr) {
std::rethrow_exception(eptr);
}
} catch(const std::runtime_error& e) {
std::cerr << "Connection failure \"" << e.what() << "\"\n";
} catch(const std::out_of_range& e) {
std::cerr << "Lease expiry \"" << e.what() << "\"\n";
}
};
etcd::KeepAlive keepalive(etcd, handler, ttl, lease_id);
Without handler, the internal state can be checked via KeepAlive::Check()
and it will rethrow
the async exception when there are errors during keeping the lease alive.
- Cancellation of asynchronous calls(except for watch)
This project is licensed under the BSD-3-Clause license - see the LICENSE.