Stability of large Hazelcast cluster - cluster-computing

We have (won't be true for too long, if the powers that be have their way) a reasonably large cluster of about 600 nodes, all of them under the same "Group Name", while only a fraction of them (about a dozen) ever made it into the list of TCP/IP Interfaces defined in the hazelcast.xml
Here's our configuration
<hazelcast xsi:schemaLocation="http://www.hazelcast.com/schema/config hazelcast-config-3.1.xsd"
xmlns="http://www.hazelcast.com/schema/config"
xmlns:xsi="http://www.w3.org/2001/XMLSchema-instance">
<group>
<name>BlappityBlah</name>
<password>blahBlaha</password>
</group>
<management-center enabled="false"/>
<network>
<port auto-increment="true">6401</port>
<outbound-ports>
<!--
Allowed port range when connecting to other nodes.
0 or * means use system provided port.
-->
<ports>0</ports>
</outbound-ports>
<join>
<multicast enabled="false">
<multicast-group>224.2.2.3</multicast-group>
<multicast-port>54327</multicast-port>
</multicast>
<tcp-ip enabled="true">
<interface>10.50.3.101-102,10.50.3.104-105,10.50.3.108-112,10.60.2.20,10.60.3.103,10.60.4.106-107</inter
face>
</tcp-ip>
<aws enabled="false">
<access-key>my-access-key</access-key>
<secret-key>my-secret-key</secret-key>
<!--optional, default is us-east-1 -->
The rest are only bound by the "Group Name", which defines the cluster, per my understanding. We don't use multicast in our configuration.
Primary application of our cluster is in distributed locking. What we are noticing of late is the arbitrary timeouts and dropping of connection between nodes, repeated "re-partitioning" and hanging locks. Everything freezes up after a while.. Earlier we ended up rebooting the nodes, now we use the Hazelcast TestApp console to clear up the map of locks.
I can vouch on the fact that the code that locks and unlocks are reasonably water tight.
My observation.. We didn't have these kind of issues earlier, until we updated Hazelcast to 3.1.5 AND scaled our nodes from 30 odd to now 500+, of which most nodes are JVMs, often up to a dozen on the same physical node. This didn't happen overnight, it was gradual.
a) Does the fact that most of our nodes don't figure in the hazelcast.xml impact their stability as members of the cluster?
b) Has anybody seen issues with scaling, is this a Hazelcast bug, or are we doing something terribly wrong while the rest of you are having a ball with Hazelcast?

a) Does the fact that most of our nodes don't figure in the hazelcast.xml impact their stability as members of the cluster?
No.
b) Has anybody seen issues with scaling, is this a Hazelcast bug, or are we doing something terribly wrong while the rest of you are having a ball with Hazelcast?
The chance cluster repartitioning increases as you add nodes. I.e. if the chance of a single node failing is e.g. 0.01% per day, then with 600 nodes, your chance of seeing a daily node failure (= repartitioning) is almost 6%.
With a chance of 0.001% failure per node per day, you'd still be around 1% chance cluster-wide.
In other words, you're cluster is probably larger than what's advisable, regardless of implementation.

Related

Is one large AWS instance better than several smaller instances for the scylladb

I have one scylla db cluster with 9 nodes and RF=3 using amazon AWS i3en.xlarge instance.
I'm curious if 3 i3en.3xlarge are much better than 9 i3en.xlarge.
Full disclosure - I work on the ScyllaDB project.
Theoretically, Scylla's shard-per-core architecture means that 16 4xlarges or 4 16xlarges should perform fundamentally the same. Each vCPU performs as in independent shared-nothing shard doing its own thing. So, how those shards are configured is irrelevant.
However, in the real world, there are good reasons for scaling up, rather than scaling out. For example:
Larger nodes have better network guarantees from AWS.
Larger nodes have fewer noisy neighbor problems.
Managing a few nodes is generally easier than managing many nodes.
Generally speaking, our users have had better experiences with larger nodes. But the choice is yours.

Consul support or alternative for 2 nodes

I want to use consul for a 2-node cluster. Drawback is there's no failure tolerance for two nodes :
https://www.consul.io/docs/internals/consensus.html
Is there a way in Consul to make a consistent leader election with only two nodes? Can Consul Raft Consensus algorithm be changed?
Thanks a lot.
It sounds like you're limited to 2 machines of this type, because they are expensive. Consider acquiring three or five cheaper machines to run your orchestration layer.
To answer protocol question, no, there is no way to run a two-node cluster with failure tolerance in Raft. To be clear, you can safely run a two-node cluster just fine - it will be available and make progress like any other cluster. It's just when one machine goes down, because your fault tolerance is zero you will lose availability and no longer make no progress. But safety is never compromised - your data is still persisted consistently on these machines.
Even outside Raft, there is no way to run a two-node cluster and guarantee progress upon a single failure. This is a fundamental limit. In general, if you want to support f failures (meaning remain safe and available), you need 2f + 1 nodes.
There are non-Raft ways to improve the situation. For example, Flexible Paxos shows that we can require both nodes for leader election (as it already is in Raft), but only require a single node for replication. This would allow your cluster to continue working in some failure cases where Raft would have stopped. But the worst case is still the same: there are always failures that will cause any two-node cluster to become unavailable.
That said, I'm not aware of any practical flexible paxos implementations anyway.
Considering the expense of even trying to hack up a solution to this, your best bet is to either get a larger set of cheaper machines, or just run your two-node cluster and accept unavailability upon failure.
Talking about changing the protocol, there is impossibility proof by FLP which states that consensus cannot be reached if systems are less than 2f + 1 for f failures (fail-stop). Although, safety is provided but progress (liveness) cannot be ensured.
I think, the options suggested in earlier post are the best.
The choice of leader election on top of the Consul’s documentation itself requires 3 nodes. This relies on the health-checks mechanism, as well as the sessions. Sessions are essentially distributed locks automatically released by TTL or when the service crashes.
To build 2-node Consul cluster we have to use another approach, supposedly called Leader Lease. Since we already have Consul KV-storage with CAS support, we can simply write to it which machine is the leader before the expiration of such and such time. As long as the leader is alive and well, it can periodically extend it's time. If the leader dies, someone will replace it quickly. For this approach to work, it is enough to synchronize the time on the machines using ntpd and when the leader performs any action, verify that it has enough time left to complete this action.
A key is created in the KV-storage, containing something like “node X is the leader before time Y”, where Y is calculated as the current time + some time interval(T). As a leader, node X updates the record once every T/2 or T/3 units of time, thereby extending it's leadership role. If a node falls or cannot reach the KV-storage, after the interval(T) its place will be taken by the node, which will be the first to discover that the leadership role has been released.
CAS is needed to prevent a race condition if the two nodes simultaneously try to become a leader. CAS Specifies to use a Check-And-Set operation. This is very useful as a building block for more complex synchronization primitives. If the index is 0, Consul will only put the key if it does not already exist. If the index is non-zero, the key is only set if the index matches the ModifyIndex of that key.

Infinispan vs memcached for high concurrency need

My web application maintains in memory cache of domain entities which are read/written at high frequency. To make application clustered, i need to synchronize / externalize this cache.
Which will be better option amongst memcached and infinispan considering following application facts-
cache will be read/written at high frequency per second
if infinispan, data need to replicated across nodes near- real time
high concurrent write should not create conflicts issue if replication is slow.
I feel memcached will solve this purpose well since it's centralized and does not need replication delay like infinispan. Can experts provide opinion on this?
Unfortunately I'm not a Memcached expert but let me tell you more about some fundamental concepts so that you could pick the best option for your use case...
First, centralized vs decentralized - if you have only one node in your system, it will be faster (as you said there is no replication). However what will happen if the node is down? Or another scenario - what will happen if the node gets full (as you said you will perform a lot of read/writes per second)? One solution for that is to use master/slave replication where writes are propagated to the slave node asynchronously. This solution will save you in case the node is down but won't do any good if the node is full (if master node is full, slave will get full a couple of minutes later).
Data consistency - if you have more than 1 node in your system, your data might get out of sync. Imagine asynchronous replication between 2 nodes and a client connected to each of them. Both clients perform a write to the same key at the same exact moment. It might seems unlikely but believe me, with highly concurrent reads and writes it will happen. The only way to solve this problem is to use synchronous replication with majority of nodes up and running (or with so called consensus).
Back to your scenario - if a broken node is not a problem for you (for example, you can switch to some other data source automatically) and your data won't grow - go ahead for 1 node solution or master/slave replication. If your data need to be strongly consistent - make sure you're doing sync replication (and possibly with transactions but you need to refer to the user manual for guidance). Otherwise I would recommend picking a more versatile solution which will allow you to add/remove nodes without taking down whole system and will have an option for sync/async replication.
From my experience, people care too much about data consistency whereas should care much more about scalability. And a final piece of advice - please define your performance criteria before evaluating any solution (something like, my writes need to take no longer than X and reads no longer than Y. Define also confidence level for your criteria (I need 99.5% of all reads to be less than X).

optimize elasticsearch / JVM

I have a website for classified. For this I'm using elasticsearch, postgres and rails on a same ubuntu 14.04 dedicated server, with 256GB of RAM and 20 cores, 40 threads.
I have 10 indexes on elasticsearch, each have default numbers of shards (5). They have between 1000 and 400 000 classifieds depending on which index.
approximately 5000 requests per minute, 2/3 making an elasticsearch request.
according to htop, jvm is using around 500% of CPU
I try different options, I reduce number of shards per index, I also try to change JAVA_OPTS as followed
#JAVA_OPTS="$JAVA_OPTS -XX:+UseParNewGC"
#JAVA_OPTS="$JAVA_OPTS -XX:+UseConcMarkSweepGC"
#JAVA_OPTS="$JAVA_OPTS -XX:CMSInitiatingOccupancyFraction=75"
#JAVA_OPTS="$JAVA_OPTS -XX:+UseCMSInitiatingOccupancyOnly"
JAVA_OPTS="$JAVA_OPTS -XX:+UseG1GC"
but it doesn't seems to change anything.
so to questions :
when you change any setting on elasticsearch, and then restart, should the improvement (if any) be visible immediately or can it arrive a bit later thanks to cache or anything else ?
can any one help me to find good configuration for JVM / elasticsearch so it will not take that many resources
First, it's a horrible idea to run your web server, database and Elasticsearch server all on the same box. Each of these should be given it's own box, at least. In the case of Elasticsearch, it's actually recommended to have at least 3 servers, or nodes. That way you end up with a load balanced cluster that won't run into split-brain issues.
Further, sharding only makes sense in a cluster. If you only have one node, then all the shards reside on the same node. This causes two performance problems. First, you get the hit that sharding always adds. For every query, Elasticsearch must query each shard individually (each is a separate Lucene index). Then, it must combine and process the result from all the shards to produce the final result. That's a not insignificant amount of overhead. Second, because all the shards reside on the same node, you're I/O-locked. The shards have to be queried one at a time instead of all at once. Optimally, you should have one shard per node, however, since you can't create more shards without reindexing, it's common to have a few extra hanging around for future horizontal scaling. In that scenario, the cost of reindexing what could be 100's of gigs of data or more outweighs a little bit of performance bottleneck. However, if you've got 5 shards running one node, that's probably a large part of your performance problems right there.
Finally, and again, with Elasticsearch in particular, swapping is a huge no-no. Most of what makes Elasticsearch efficient is it's cache which all resides in RAM. If swaps occur, it jacks with the cache in sometimes unpredictable ways. As result, it's recommended to turn off swapping completely on the box your node(s) run on, and set Elasticsearch/JVM to have a min and max memory consumption of roughly half the available RAM of the box. That's virtually impossible to achieve if you have other things running on it like a web server or database. Databases in particular aggressively consume RAM in order to increase throughput, which is why those should likewise reside on their own servers.

Block Replication Limits in HDFS

I'm currently rebuilding our servers that have our region-servers and data nodes. When I take down a data node, after 10 minutes the blocks that it had are being re-replicated among other data nodes, as it should. We have 10 data-nodes, so I see heavy network traffic as the blocks are being re-replicated. However, I'm seeing that traffic to be about only 500-600mbps per server (the machines all have gigabit interfaces) so it's definitely not network-bound. I'm trying to figure out what is limiting the speed that the data-nodes send and receive blocks. Each data-node has six 7200 rpm sata drives, and the IO usage is very low during this, only peaking to 20-30% per drive. Is there a limit built into hdfs that limits the speed at which blocks are replicated?
The rate of replication work is throttled by HDFS to not interfere with cluster traffic when failures happen during regular cluster load.
The properties that control this are dfs.namenode.replication.work.multiplier.per.iteration (2), dfs.namenode.replication.max-streams (2) and dfs.namenode.replication.max-streams-hard-limit (4). The foremost controls the rate of work to be scheduled to a DN at every heartbeat that occurs, and the other two further limit the maximum parallel threaded network transfers done by a DataNode at a time. The values in () indicate their defaults. Some description of this is available at https://hadoop.apache.org/docs/current/hadoop-project-dist/hadoop-hdfs/hdfs-default.xml
You can perhaps try to increase the set of values to (10, 50, 100) respectively to spruce up the network usage (requires a NameNode restart), but note that your DN memory usage may increase slightly as a result of more blocks information being propagated to it. A reasonable heap size for these values for the DN role would be about 4 GB.
P.s. These values were not tried by me on production systems personally. You will also not want to max out the re-replication workload such that it affects regular cluster work, as recovery of 1/3 replicas may be of lesser priority than missing job/query SLAs due to lack of network resources (unless you have a really fast network that's always under-utilised even under loaded periods). Try to tune it till you're satisfied with the results.

Resources