Hi Bart,
Max from ArangoDB here. We are putting the finishing touches on the 3.0
release basically as we speak. Good cluster setup documentation is
unfortunately one of the things we are still missing, but we will
definitely have this in place for the release.
Am Mittwoch, 25. Mai 2016 22:49:17 UTC+2 schrieb Bart DS:
>
> Hi Willi,
>
> Thanks for this very detailed explanation.
> If I understand correctly, I can set up an ArangoDB 3.0 cluster without
> any orchestration framework (such as Mesos) ?
>
Yes, the startup process has been simplified a lot for 3.0. You basically
have to fire up a bunch of Docker containers (all with the same image) with
certain command line options. Everything else organises itself within the
ArangoDB cluster. We will publish a blog post shortly after the release to
explain how this is done.
> When using asynchronous replication there will be no automatic failover or
> rebalancing (yet), but when using synchronous replication I will have
> automatic failover and rebalancing out of the box?
>
Yes, all of this is done within the ArangoDB cluster thanks to our own
implementation of the Raft consensus protocol. The only thing it cannot do
on its own is restart containers or launch new ones. And automatic failover
for the asynchronous replication within the ArangoDB cluster will only land
with 3.1.
>
> So if I have e.g. 5 servers in the cluster and one server goes down, one
> (or more?) of the other 4 servers will take over the tasks of the failing
> server and all reads/writes will still succeed?
> Is that correct?
>
Exactly so. And due to synchronous replication you will not lose committed
and confirmed data.
>
> Also, If I run 3.0 on another orchestration framework, will I be able to
> dynamically add/remove servers to/from the cluster already? Or isn't that
> possible yet?out
>
You can simply start new Docker containers to add more coordinators or
DBServers. You can simply kill coordinators without losing anything.
You can ask the ArangoDB cluster to clean out a DBserver in a controlled
fashion such that its data is relocated to other servers automatically.
Once this has completed, you can simply kill the DBServer by stopping its
Docker container.
>
> Can you point me to any documentation on how to set up such a cluster with
> ArangoDB 3.0?
>
Unfortunately, this is not yet written. I do have a bash script which
launches a cluster locally just using docker containers. I attach the
script such that you can see how things will work. This uses the 3.0.0b3
which we are about to publish. Note that in this not all of the above
mentioned features work yet.
Cheers,
Max.
>
> Bart
>
--
You received this message because you are subscribed to the Google Groups
"ArangoDB" group.
To unsubscribe from this group and stop receiving emails from it, send an email
to [email protected].
For more options, visit https://groups.google.com/d/optout.
#!/bin/bash
if [ -z "$XTERM" ] ; then
XTERM=x-terminal-emulator
fi
if [ -z "$XTERMOPTIONS" ] ; then
XTERMOPTIONS="--geometry=80x43"
fi
NRDBSERVERS=$1
if [ "$NRDBSERVERS" == "" ] ; then
NRDBSERVERS=2
fi
echo Number of DBServers: $NRDBSERVERS
NRCOORDINATORS=$2
if [ "$NRCOORDINATORS" == "" ] ; then
NRCOORDINATORS=1
fi
echo Number of Coordinators: $NRCOORDINATORS
if [ ! -z "$3" ] ; then
if [ "$3" == "C" ] ; then
COORDINATORCONSOLE=1
echo Starting one coordinator in terminal with --console
fi
fi
SECONDARIES="$4"
echo Starting agency...
docker run -d --net=host -e ARANGO_NO_AUTH=1 --name=agency \
arangodb/arangodb-preview:3.0.0b1 \
--agency.endpoint tcp://0.0.0.0:4001 \
--agency.id 0 \
--agency.size 1 \
--agency.wait-for-sync false \
--agency.supervision false \
--agency.supervision-frequency 5 \
--server.endpoint tcp://0.0.0.0:4001 \
--server.statistics false \
--server.threads 16
sleep 1
start() {
if [ "$1" == "dbserver" ]; then
ROLE="PRIMARY"
elif [ "$1" == "coordinator" ]; then
ROLE="COORDINATOR"
fi
TYPE=$1
PORT=$2
echo Starting $TYPE on port $PORT
docker run -d --net=host -e ARANGO_NO_AUTH=1 --name="$TYPE_$PORT" \
arangodb/arangodb-preview:3.0.0b1 \
--cluster.agency-endpoint tcp://127.0.0.1:4001 \
--cluster.my-address tcp://127.0.0.1:$PORT \
--server.endpoint tcp://127.0.0.1:$PORT \
--cluster.my-local-info $TYPE:127.0.0.1:$PORT \
--cluster.my-role $ROLE \
--log.level trace \
--server.statistics false
}
startTerminal() {
if [ "$1" == "dbserver" ]; then
ROLE="PRIMARY"
elif [ "$1" == "coordinator" ]; then
ROLE="COORDINATOR"
fi
TYPE=$1
PORT=$2
echo Starting $TYPE on port $PORT
$XTERM $XTERMOPTIONS -e docker run -it --net=host -e ARANGO_NO_AUTH=1 \
--name="$TYPE_$PORT" \
arangodb/arangodb-preview:3.0.0b1 \
--cluster.agency-endpoint tcp://127.0.0.1:4001 \
--cluster.my-address tcp://127.0.0.1:$PORT \
--server.endpoint tcp://127.0.0.1:$PORT \
--cluster.my-local-info $TYPE:127.0.0.1:$PORT \
--cluster.my-role $ROLE \
--server.statistics false \
--console &
}
PORTTOPDB=`expr 8629 + $NRDBSERVERS - 1`
for p in `seq 8629 $PORTTOPDB` ; do
start dbserver $p
done
PORTTOPCO=`expr 8530 + $NRCOORDINATORS - 1`
for p in `seq 8530 $PORTTOPCO` ; do
if [ $p == "8530" -a ! -z "$COORDINATORCONSOLE" ] ; then
startTerminal coordinator $p
else
start coordinator $p
fi
done
echo Waiting for cluster to come up...
testServer() {
PORT=$1
while true ; do
curl -s -f -X GET "http://127.0.0.1:$PORT/_api/version"; > /dev/null 2>&1
if [ "$?" != "0" ] ; then
echo Server on port $PORT does not answer yet.
else
echo Server on port $PORT is ready for business.
break
fi
sleep 1
done
}
for p in `seq 8629 $PORTTOPDB` ; do
testServer $p
done
for p in `seq 8530 $PORTTOPCO` ; do
testServer $p
done
echo Done, your cluster is ready at
for p in `seq 8530 $PORTTOPCO` ; do
echo " build/bin/arangosh --server.endpoint tcp://127.0.0.1:$p"
done
