For an ActiveMQ Artemis HA solution, I'm thinking that instead of using a using a primary/secondary configuration with a Zookeeper monitoring node, perhaps as master/slave/slave configuration would be better. The reason is that, as pointed out in this answer, having a single Zookeeper node is an architectural weakness. The master-slave-slave approach would at least probably have at least one node running at any given time.
Also, I'm wondering if it might be better to ask these question over the mailing list or the slack channel? Any preference there?
Using a primary/backup/backup triplet is possible, but it provides no mitigation against split brain since only primary nodes participate in quorum voting. Therefore, the architectural weakness in this configuration is significantly worse than just have a single ZooKeeper node because with a single ZooKeeper node at least you have a chance of mitigating split brain whereas otherwise you don't.
Also, fail-back only works for a primary/backup pair. It doesn't work for a primary/backup/backup triplet. This is because a backup server is owned by only one primary server. This means that when the 3 broker instances are started there will be 1 primary/backup pair and a "left-over" backup which will be in a kind of idle state waiting to attach to a primary broker without a backup. Then if the primary broker fails the primary broker's backup will take over and become primary and the other backup will now become the backup of the server which just became primary. Once the broker instance which failed is restarted it will attempt to register itself as a backup of the now-primary broker and initiate fail-back. However, since the now-primary broker already has a backup it will reject the registration message from the original primary because it already has a backup, and therefore fail-back will not occur.
Related
We have a small Elasticsearch cluster for 3 nodes: two in one datacenter and one in another for disaster recovery reasons. However, if the first two nodes fail simultaneously, the third one won't work either - it will just throw "master not discovered or elected yet".
I understand that this is intended - this is how Elasticsearch cluster should work. But is there some additional special configuration that I don't know to keep the third single node working, even if in the read-only mode?
nope, there's not. as you mentioned it's designed that way
you're probably not doing yourselves a lot of favours by running things across datacentres like that. network issues are not kind on Elasticsearch due to it's distributed nature
Elasticsearch runs in distributed mode by default. Nodes assume that there are or will be a part of the cluster, and during setup nodes try to automatically join the cluster.
If you want your Elasticsearch to be available for only node without the need to communicate with other Elasticsearch nodes. It works similar to a standalone server. To do this we can tell Elasticsearch to work in local only (disable network)
open your elasticsearch/config/elasticsearch.yml and set:
node.local: true
After follow the guide below, i manage to set up a active-passive cluster node. But i notice that when the main fails and come back on, it does not take back the primary role. What setting i should configure in the crm to ensure that when the primary recover after a fail it takes back as primary from the back up machine?
https://www.theurbanpenguin.com/drbd-pacemaker-ha-cluster-ubuntu-16-04/
You would do this via an infinity location constraint. Something like:
location fs-on-alice fs_res inf: alice
Automatic fail-back is usually not suggested. In an ideal HA configuration it shouldn't matter at all what system is currently primary. Also, I have seen instances where a node is having intermittent panics/reboots every 10-20 minutes. Now you have services failing over, and stopping/restarting, several times an hour. If not for the location constraint and automatic fail-backs you would not have this behavior.
well giving a "inf" location rule will always force the resource to run on alice only. And if alice goes down for maintenance then the resource will not run on any other node. What u want to add is called stickiness.
pcs constraint location fs_res prefers alice =50
Refer for more detail: http://clusterlabs.org/pacemaker/doc/en-US/Pacemaker/1.1/html/Clusters_from_Scratch/_prefer_one_node_over_another.html
I am new to the topic. Having read a handful of articles on it, and asked a couple of persons, I still do not understand what you people do in regard to one problem.
There are UI clients making requests to several backend instances (for now it's irrelevant whether sessions are sticky or not), and those instances are connected to some highly available DB cluster (may it be Cassandra or something else of even Elasticsearch). Say the backend instance is not specifically tied to one or cluster's machines, and instead its every request to DB may be served by a different machine.
One client creates some record, it's synchronously of asynchronously stored to one of cluster's machines then eventually gets replicated to the rest of DB machines. Then another client requests the list or records, the request ends up served by a distant machine not yet received the replicated changes, and so the client does not see the record. Well, that's bad but not yet ugly.
Consider however that the second client hits the machine which has the record, displays it in a list, then refreshes the list and this time hits the distant machine and again does not see the record. That's very weird behavior to observe, isn't it? It might even get worse: the client successfully requests the record, starts some editing on it, then tries to store the updates to DB and this time hits the distant machine which says "I know nothing about this record you are trying to update". That's an error which the user will see while doing something completely legitimate.
So what's the common practice to guard against this?
So far, I only see three solutions.
1) Not actually a solution but rather a policy: ignore the problem and instead speed up the cluster hard enough to guarantee that 99.999% of changes will be replicated on the whole cluster in, say, 0.5 secord (it's hard to imagine some user will try to make several consecutive requests to one record in that time; he can of course issue several reading requests, but in that case he'll probably not notice inconsistency between results). And even if sometimes something goes wrong and the user faces the problem, well, we just embrace that. If the loser gets unhappy and writes a complaint to us (which will happen maybe once a week or once an hour), we just apologize and go on.
2) Introduce an affinity between user's session and a specific DB machine. This helps, but needs explicit support from the DB, and also hurts load-balancing, and invites complications when the DB machine goes down and the session needs to be re-bound to another machine (however with proper support from DB I think that's possible; say Elasticsearch can accept routing key, and I believe if the target shard goes down it will just switch the affinity link to another shard - though I am not entirely sure; but even if re-binding happens, the other machine may contain older data :) ).
3) Rely on monotonic consistency, i.e. some method to be sure that the next request from a client will get results no older than the previous one. But, as I understand it, this approach also requires explicit support from DB, like being able so pass some "global version timestamp" to a cluster's balancer, which it will compare with it's latest data on all machines' timestamps to determine which machines can serve the request.
Are there other good options? Or are those three considered good enough to use?
P.S. My specific problem right now is with Elasticsearch; AFAIK there is no support for monotonic reads there, though looks like option #2 may be available.
Apache Ignite has primary partition for a key and backup partitions. Unless you have readFromBackup option set, you will always be reading from primary partition whose contents is expected to be reliable.
If a node goes away, a transaction (or operation) should be either propagated by remaining nodes or rolled back.
Note that Apache Ignite doesn't do Eventual Consistency but instead Strong Consistency. It means that you can observe delays during node loss, but will not observe inconsistent data.
In Cassandra if using at least quorum consistency for both reads and writes you will get monotonic reads. This was not the case pre 1.0 but thats a long time ago. There are some gotchas if using server timestamps but thats not by default so likely wont be an issue if using C* 2.1+.
What can get funny is since C* uses timestamps is things that occur at "same time". Since Cassandra is Last Write Wins the times and clock drift do matter. But concurrent updates to records will always have race conditions so if you require strong read before write guarantees you can use light weight transactions (essentially CAS operations using paxos) to ensure no one else updates between your read to update, these are slow though so I would avoid it unless critical.
In a true distributed system, it does not matter where your record is stored in remote cluster as long as your clients are connected to that remote cluster. In Hazelcast, a record is always stored in a partition and one partition is owned by one of the servers in the cluster. There could be X number of partitions in the cluster (by default 271) and all those partitions are equally distributed across the cluster. So a 3 members cluster will have a partition distribution like 91-90-90.
Now when a client sends a record to store in Hazelcast cluster, it already knows which partition does the record belong to by using consistent hashing algorithm. And with that, it also knows which server is the owner of that partition. Hence, the client sends its operation directly to that server. This approach applies on all client operations - put or get. So in your case, you may have several UI clients connected to the cluster but your record for a particular user is stored on one server in the cluster and all your UI clients will be approaching that server for their operations related to that record.
As for consistency, Hazelcast by default is strongly consistent distributed cache, which implies that all your updates to a particular record happen synchronously, in the same thread and the application waits until it has received acknowledgement from the owner server (and the backup server if backups are enabled) in the cluster.
When you connect a DB layer (this could be one or many different types of DBs running in parallel) to the cluster then Hazelcast cluster returns data even if its not currently present in the cluster by reading it from DB. So you never get a null value. On updating, you configure the cluster to send the updates downstream synchronously or asynchronously.
Ah-ha, after some even more thorough study of ES discussions I found this: https://www.elastic.co/guide/en/elasticsearch/reference/current/search-request-preference.html
Note how they specifically highlight the "custom value" case, recommending to use it exactly to solve my problem.
So, given that's their official recommendation, we can summarise it like this.
To fight volatile reads, we are supposed to use "preference",
with "custom" or some other approach.
To also get "read your
writes" consistency, we can have all clients use
"preference=_primary", because primary shard is first to get all
writes. This however will probably have worse performance than
"custom" mode due to no distribution. And that's quite similar to what other people here said about Ignite and Hazelcast.
Right?
Of course that's a solution specifically for ES. Reverting to my initial question which is a bit more generic, turns out that options #2 and #3 are really considered good enough for many distributed systems, with #3 being possible to achieve with #2 (even without immediate support for #3 by DB).
Just to make the scenario simple.
number of consumers == number of partitions == Kafka broker numbers
If deploy the consumers on the same machines where the brokers are, how to make each consumer only consume the messages locally? The purpose is to cut all the network overhead.
I think we can make it if each consumer can know the partition_id on their machines, but I don't know how? or is there other directions to solve this problem?
Thanks.
bin/kafka-topics.sh --zookeeper [zk address] --describe --topic [topic_name] tells you which broker hosts the leader for each partition. Then you can use manual partition assignment for each consumer to make sure it consumes from a local partition.
Probably not worth the effort because partition leadership can change and then you would have to rebalance all your consumers to be local again. You can save the same amount of network bandwidth with less effort by just reducing the replication factor from 3 to 2.
Maybe you could use the Admin Client API.
First you can use the describeTopics() methods for getting information about topics in the cluster. From the DescribeTopicResult you can access to TopicPartitionInfo with information about partitions for each topic. From there you can access to the Node through the leader(). Node contains the host() and you can check if it's the same as the host your consumer is running or id() and the consumer should have the information about the broker-id running on the same machine (in general it's an information you can define upfront). More infor on Admin Client API at the following JavaDoc :
https://kafka.apache.org/0110/javadoc/index.html?org/apache/kafka/clients/admin/AdminClient.html
After going through H2 developer guide I still don't understand how can I find out what cluster node(s) was/were failing and which database needs to be recovered in the event of temporary network failure.
Let's consider the following scenario:
H2 cluster started with N active nodes (is actually it true that H2 can support N>2, i.e. more than 2 cluster nodes?)
(lots DB updates, reads...)
Network connection with one (or several) cluster nodes gets down and node becomes invisible to the rest of the cluster
(lots of DB updates, reads...)
Network link with previously disconnected node(s) restored
It is discovered that cluster node was probably missing (as far as I can see SELECT VALUE FROM INFORMATION_SCHEMA.SETTINGS WHERE NAME='CLUSTER' starts responding with empty string if one node in cluster fails)
After this point it is unclear how to find out what nodes were failing?
Obviously, I can do some basic check like comparing DB size, but it is unreliable.
What is the recommended procedure to find out what node was missing in the cluster, esp. if query above responds with empty string?
Another question - why urlTarget doesn't support multiple parameters?
How I am supposed to use CreateCluster tool if multiple nodes in the cluster failed and I want to recover more than one?
Also I don't understand how CreateCluster works if I had to stop the cluster and I don't want to actually recover any nodes? What's not clear to me is what I need to pass to CreateCluster tool if I don't actually need to copy database.
That is partially right SELECT VALUE FROM INFORMATION_SCHEMA.SETTINGS WHERE NAME='CLUSTER', will return an empty string when queried in standard mode.
However, you can get the list of servers by using Connection.getClientInfo() as well, but it is a two-step process. Paraphrased from h2database.com:
The list of properties returned by getClientInfo() includes a numServers property that returns the number of servers that are in the connection list. getClientInfo() also has properties server0..serverN, where N is the number of servers - 1. So to get the 2nd server from the list you use getClientInfo('server1').
Note: The serverX property only returns IP addresses and ports and not
hostnames.
And before you say simple replication, yes that is default operation, but you can do more advanced things that are outside the scope of your question in clustered H2.
Here's the quote for what you're talking about:
Clustering can only be used in the server mode (the embedded mode does not support clustering). The cluster can be re-created using the CreateCluster tool without stopping the remaining server. Applications that are still connected are automatically disconnected, however when appending ;AUTO_RECONNECT=TRUE, they will recover from that.
So yes if the cluster stops, auto_reconnect is not enabled, and you stick with the basic query, you are stuck and it is difficult to find information. While most people will tell you to look through the API and or manual, they haven't had to look through this one so, my sympathies.
I find it way more useful to track through the error codes, because you get a real good idea of what you can do when you see how the failure is planned for ... here you go.