Running couchbase on AWS under auto scaling - couchbase

I want to run couchbase on AWS EC2. Since my traffic is cyclic in nature, can I run Couchbase under auto-scaling. Since there are a lot of steps required to add/remove a node, I was wondering if this is the right approach. Has anybody tried it ?

It has been done before. Here is a high level list of the things you'd have to do:
Define which Couchbase metrics you need to use to base your scaling considerations on
Create a script to get those metrics from Couchbase and put them into Cloudwatch using Couchbase Rest API or CLI.
Create an AMI with Couchbase installed and OS configured.
Script the addition of one or more new nodes (using Couchbase Rest API or CLI), plus a rebalance, as a response to Auto-scaling
Script the removal and rebalance of nodes (using Couchbase Rest API or CLI), as a response to contraction in auto-scaling.
With you reliying on rebalances here, you will have to watch how long your rebalances take and perhaps tune your cluster (e.g. move more vBuckets at once and other settings) and usage of Couchbase for faster rebalances (e.g. if you have large views, they can have an effect on rebalances). Normally rebalances are meant to be a background process and take as long as they take, but that may not be appropriate in this particular use. Only you can answer that.

Related

Using Couchbase SDK vs Sync Gateway API

I have a full deployment of couchbase (server, sync gateway and lite) and have an API, mobile app and web app all using it.
It works very well, but I was wondering if there are any advantages to using the Sync Gateway API over the Couchbase SDK? Specifically I would like to know if Sync Gateway would handle larger numbers of operations better than the SDK, perhaps an internal queue/cache system, but can't seem to find definitive documentation for this.
At the moment the API uses the C# Couchbase SDK and we use SyncGateway very little (only really for synchronising the mobile app).
First, some relevant background info :
Every document that needs to be synced over to Couchbase Lite(CBL) clients needs to be processed by the Sync Gateway (SGW). This is true whether a doc is written via the SGW API or whether it comes in via server write (N1QL or SDK). The latter case is referred to as "import processing” wherein the document that is written to the bucket (via N1QL) is read by SGW via DCP feed. The document is then processed by SGW and written back to the bucket with the relevant sync metadata.
Prerequisite :
In order for the SGW to import documents written directly via N1QL/SDK, you must enable “shared bucket access” and import processing as discussed here
Non-mobile documents :
If you have documents that are never going to be synced to the CBL clients, then choice is obvious. Use server SDKs or N1QL
Mobile documents (docs to sync to CBL clients) :
Assuming you are on SGW 2.x syncing with CBL 2.x clients
If you have documents written at server end that need to be synced to CBL clients, then consider the following
Server side write rate:
If you are looking at writes on server side coming in at sustained rates significantly exceeding 1.5K/sec (lets say 5K/sec), then you should go the SGW API route. While it's easy enough to do a bulk update via server N1QL query, remember that SGW still needs to keep up and do the import processing (what's discussed in the background).
Which means, if you are doing high volume updates through the SDK/N1QL, then you will have to rate limit it so the SGW can keep up (do batched updates via SDK)
That said, it is important to consider the fact that if SGW can't keep up with the write throughput on the DCP feed, it's going to result in latency, no matter how the writes are happening (SGW API or N1QL)
If your sustained write rate on server isn’t excepted to be significantly high, then go with N1QL.
Deletes Handling:
Does not matter. Under shared-bucket-access, deletes coming in via SDK or SGW API will result in a tombstone. Read more about it here
SGW specific config :
Naturally, if you are dealing with SGW specific config, creating SGW users, roles, then you will use the SGW API for that.
Conflict Handling :
In 2.x, it does not matter. Conflicts are handled on CBL side.
Challenge with SGW API
Probably the biggest challenge in a real-world scenario is that using the SG API path means either storing information about SG revision IDs in the external system, or perform every mutation as a read-then-write (since we don't have a way to PUT a document without providing a revision ID)
The short answer is that for backend operations, Couchbase SDK is your choice, and will perform much better. Sync Gateway is meant to be used by Mobile clients, with few exceptions (*).
Bulk/Batch operations
In my performance tests using Java Couchbase SDK and bulk operations from AsyncBucket (link), I have updated up to 8 thousand documents per second. In .Net there you can do Batch operations too (link).
Sync Gateway also supports bulk operations, yet it is much slower because it relies on REST API and it requires you to provide a _rev from the previous version of each document you want to update. This will usually result in the backend having to do a GET before doing a PUT. Also, keep in mind that Sync Gateway is not a storage unit. It just works as a proxy to Couchbase, managing mobile client access to segments of data based on the channels registered for each user, and writes all of it's meta-data documents into the Couchbase Server bucket, including channel indexing, user register, document revisions and views.
Querying
Views are indexed thus for querying of large data they may will respond very fast. Whenever a document is changed, the map function of all views has the opportunity to map it. But when a view is created through Sync Gateway REST API, some code is added to your map function to handle user channels/permissions, making it slower than plain code created directly in Couchbase Admin UI. Querying views with compound keys using startKey/endKey parameters is very powerful when you have hierarchical data, but this functionality and the use of reduce function are not available for mobile clients.
N1QL can also be very fast too, when your N1QL query is taking advantage of Couchbase indexes.
Notes
(*) One exception to the rule is when you want to delete a document and have this reflected on mobile phones. The DELETE operation, leaves an empty document with _deleted: true attribute, and can only be done through Sync Gateway. Next time the mobile device synchronizes and finds this hint, it will delete the document from local storage. You can also use set this attribute through a PUT operation, when you may also adding _exp: "2019-12-12T00:00:00.000Z" attribute to perform a programmed purge of the document in a future date, so that the server also gets clean. However, just purging a document through Sync Gateway is equivalent to delete it through Couchbase SDK and this won't reflect on mobile devices.
NOTE: Prior to Sync Gateway 1.5 and Couchbase 5.0, all backend operations had to be done directly in Sync Gateway so that Sync Gateway and mobile clients could detect those changes. This has changed since shared_bucket_access option was introduced. More info here.

Replicating all couchbase buckets from one cluster to another using XDCR

Two questions:
1) Is there any way to setup a bidirectional replication of all couchbase buckets between two clusters (without manually creating a replication between each of the buckets)
2) I'm creating buckets dynamically using the Java SDK, I would like to create and replicate a bucket created on a cluster automatically onto the other, is there anyway to do this using the Java SDK?
There is not presently a trivial way to establish bi-directional replication for all buckets on one cluster to corresponding buckets on another cluster. Nor is there a programatic Java API for managing XDCR replications.
There is however a REST API for managing XDCR replications that could be used to programatically create replications. With a little bit of work using other aspects of the REST API this could be made to provide a solution for your first question also.
The couchbase-cli tool is a command-line friendly wrapper around a significant portion of the REST API and it should be possible to create shell scripts using it to accomplish what you require if you are adverse to interacting with the REST API directly.

Couchbase SDK and Moxi Client

I am new to Couchbase and trying to understand why we need a Client side proxy like Moxi if using a Couchbase PHP SDK.
As per my understanding the proxying of client requests to right server is done by the Client SDK which maintains the vBucket map of all keys.
Why in case of a web application using PHP SDK and Couchbase, we need an additional Moxi client?
They are for two different things.
Moxi is for when you want to use a standard memcached library as MOXI will proxy memcached calls to the Couchbase cluster and use Couchbase buckets. Your code will not know it is talking to a persistent database in the background. Using moxi with Couchbase buckets will give you some of the benefits of Couchbase, like High Availability, easy scalability and performance Couchbase is known for, but you can use any old off the shelf memcached library. Just know that because of adhering to memcached, moxi is limited to that functionality from an application perspective.
In my opinion, moxi should be used to bridge the gap between people on memcached and using the full SDKs and is not meant to be a final destination, though some people have been on it for years.
Using the Couchbase PHP SDK on the other hand gives you the full suite functionality that Couchbase can offer and you do not need MOXI at all.
In summary, if you are in a spot to use the Couchbase SDK, do that. You will get more functionality, performance, etc. from it. Moxi is for those that are already have memcached, but want to step up to a clustered high performance cache and not change their code.

Heterogeneous Couchbase Cluster

I would like to run a couchbase cluster on a hardware cluster that's not uniform. Some of the machines have 1 CPU core, while the others have 16 cores.
Is there a way to configure the bucket size or request frequency so the larger servers can receive a larger percentage of the load?
What I'm looking for is something similar to the weighting in ketama, but for Couchbase.
You stated in a previous answer that:
Usually 1 small instance and 4-20 large ones. The small one basically only exists for cluster discovery.
What you should really do is connecting to your Couchbase cluster through a reverse-proxy (such as Haproxy) instead of a constantly-up node. The reverse-proxy will have all the potential nodes in its pool, constantly looking for which nodes are really up, and dispatching connections to those nodes. As soon as a node goes down, the connection will be re-established to an alive node.
You can read more about this architecture description on the Couchbase documentation.
No, there isn't method that does what you want. Keys in CB throws its own hash-function mapped to VBuckets, and VBuckets mapped to server. Couchbase API doesn't allow managed this mapping. All you can do is determined by the id of document server that owns this document.

economical way of scaling a php+mysql website

My partner and I are trying to start a website hosted in cloud. It has pretty heavy ajax traffic and the backend handles money transactions so we need ACID in some of the DB tables.
Currently everything is running off a single server. Some of the AJAX traffic are cached in text files.
Question:
What's the best way to scale the database server? I thought about moving mysql to separate instances and do master-master duplication. However this seems tough and I heard I might lose ACID properties even with InnoDB? Is Amazon RDS a good solution?
The web server is relatively stateless except for some custom log files and the ajax cache files. What's a good way to scale to multiple web servers? I guess the custom log files can be moved to a reliable shared file system or DB but not sure what to do about the AJAX cache file coherency across multiple servers. (I dont care about losing /var/log/* if web server dies)
For performance it might be cheaper to go with larger instance with more cores and memory but eventually I would need redundancy so wondering what's the best way to do this cheaply.
thanks
take a look at this post. there is plenty of presentations on the net discussing scalability. few things i suggest to keep in mind:
plan early for the data sharding [even if you are not going to do it immediately]
try using mechanisms like memcached to limit number of queries sent to the database
prepare to serve static content from other domain, in the longer run - from ngin-x-alike server and later CDN
redundancy - depends on your needs. is 'read-only' mode acceptable for your site? if so - go with mysql replication + rsync of static files and in case of failover have your site work in that mode till you recover the master node. if you need high availability - then take a look either at drbd replication [at least for mysql] or setup with automated promotion of slave server to become master node.
you might find following interesting:
http://yoshinorimatsunobu.blogspot.com/2011/08/mysql-mha-support-for-multi-master.html
http://mysqlperformanceblog.com
http://highscalability.com
http://google.com - search for scalability, lamp, failover... there are tones of case studies and horror stories from the trench lines :-]
Another option is using a scaleable platform such as Amazon Web Services. You can start out with a micro instance and configure load balancing to fire up more instances as needed.
Once you determine average resource requirements you can then resize your image to larger or smaller depending on your needs.
http://aws.amazon.com
http://tuts.pinehead.tv/2011/06/26/creating-an-amazon-ec2-instance-with-linux-lamp-stack/
http://tuts.pinehead.tv/2011/09/11/how-to-use-amazon-rds-relation-database-service-to-host-mysql/
Amazon allows you to either load balance or change instance size based off demand.