Scenario: Hundreds of nodes running an HTTP server responding to time-critical requests (requests must be processed and a response must be sent back within milliseconds eg. 50 milliseconds max). Each server will serve about 500 requests per second (making all nodes collectively serving more than 100,000 qps). In order to avoid connecting to a central (remote) database for each request, each node will have an in-memory replica of the database (database should be able to only hold a few hundred megabytes of data).
Question 1: Are there any database technologies that implement this kind of multiple (hundreds of) in-memory replicas that synchronize in realtime (or near-realtime) with a central database?
Question 2: Are there any architectural patterns that address this scenario?
Oracle's GoldenGate is a potential solution. I can't speak to the systems you tagged: redis and couchdb. I have heard positive things about redis, but it's purely anecdotal; I don't have any direct, hands-on, experience.
My own company's eXtremeDB can also accommodate this type of workload. There are a couple of configurations to choose from, but this isn't an appropriate venue to explore them with you. Please reach out to us if you'd like to get into it.
Related
I've built a REST API with a pretty decent latency. Each request is answered in ~100 ms with a thousand requests per second. This is however with a relatively low physical distance to the data center. The users of this API would, however, be spread all over the globe. From the US for example (to a data center in Germany), the response time for a single request is ~400 ms under no load.
What would be the best approach to deploying this API? I suspect multiple servers at different locations, with a load balancer in front. How would I ensure that the MySQL database would stay in sync between the servers in that case?
With multiple servers and a load balancer, the costs rise exponentially, which is something I can hopefully afford in the future, but not at the moment.
I'd love to hear your ideas!
Afaik. for big projects people use event sourcing with an event storage and microservices and message queues between them or a basic solution is polling the event storage through a simple REST API something like send me the latest events since the last event I received. If you can accept eventual consistency, then I think this approach can work pretty well. It makes writing somewhat slower, but reading can be very fast with it. No need to sync the MySQL databases directly, you just pull the latest events and use a projection to update the local MySQL database. So the event storage is the single source of truth.
I am having an application where different users may connect to different databases (those can be either MySQL or Postgres), what might be the best way to cache those connections across different databases? I saw some connection pools but seems like they are more for one db multiple connections than for multiple db multiple connections.
PS:
For adding more context, I am designing a multi tenant architecture where each tenant connects to one or multiple databases, I have an option for using map[string]*sql.DB where the key is the url of the database, but it can be hardly scaled when we have numerous number of databases. Or should we have a sharding layer for each incoming request sharded by connection url, so each machine will contain just the right amount of database connections in the form of map[string]*sql.DB?
An example for the software that I want to build is https://www.sigmacomputing.com/ where the user can connects to multiple databases for working with different tables.
Both MySQL and Postgres do not allow to connection sharing between multiple database users, single database user is specified in connection credentials. If you mean that your different users have their own database credentials, then it is not possible to share connections between them.
If by "different users" you mean your application users and if they share single database user to access DB deeper in the app, then you don't need to do anything particular to "cache" connections. sql.DB keeps and reuses open connections in its pool by default.
Go automatically opens, closes and reuses DB connections with a *database/sql.DB. By default it keeps up to 2 connections open (idle) and opens unlimited number of new connections under concurrency when all opened connections are already busy.
If you need some fine tuning on pool efficiency vs database load, you may want to alter sql.DB config with .Set* methods, for example SetMaxOpenConns.
You seem to have to many unknowns. In cases like this I would apply good, old agile and start with prototype of what you want to achieve with tools that you already know and then benchmark the performance. I think you might be surprised how much go can handle.
Since you understand how to use map[string]*sql.DB for that purpose I would go with that. You reach some limits? Add another machine behind haproxy. Solving scaling problem doesn't necessary mean writing new db pool in go. Obviously if you need this kind of power you can always do it - pgx postgres driver has it's own pool implementation so you can get your inspiration there. However doing this right now seems to be pre-mature optimization - solving problem you don't have yet. Building prototype with map[string]*sql.DB is easy, test it, benchmark it, you will see if you need more.
p.s. BTW you will most likely hit first file descriptor limit before you will be able to exhaust memory.
Assuming you have multiple users with multiple databases with an N to N relation, you could have a map of a database URL to database details (explained below).
The fact that which users have access to which databases should be handled anyway using configmap or a core database; For Database Details, we could have a struct like this:
type DBDetail {
sync.RWMutex
connection *sql.DB
}
The map would be database URL to database's details (dbDetail) and if a user is write it calls this:
dbDetail.Lock()
defer dbDetail.Unock()
and for reads instead of above just use RLock.
As said by vearutop the connections could be a pain but using this you could have a single connection or set the limit with increment and decrement of another variable after Lock.
There isn’t necessarily a correct architectural answer here. It depends on some of the constraints of the system.
I have an option for using map[string]*sql.DB where the key is the url of the database, but it can be hardly scaled when we have numerous number of databases.
Whether this will scale sufficiently depends on the expectation of how numerous the databases will be. If there are expected to be tens or hundreds of concurrent users in the near future, is probably sufficient. Often a good next step after using a map is to transition over to a more full featured cache (for example https://github.com/dgraph-io/ristretto).
A factor in the decision of whether to use a map or cache is how you imagine the lifecycle of a database connection. Once a connection is opened, can that connection remain opened for the remainder of the lifetime of the process or do connections need to be closed after minutes of no use to free up resources.
Should we have a sharding layer for each incoming request sharded by connection url, so each machine will contain just the right amount of database connections in the form of map[string]*sql.DB?
The right answer here depends on how many processing nodes are expected and whether there will be gain additional benefits from routing requests to specific machines. For example, row-level caching and isolating users from each other’s requests is an advantage that would be gained by sharing users across the pool. But a disadvantage is that you might end up with “hot” nodes because a single user might generate a majority of the traffic.
Usually, a good strategy for situations like this is to be really explicit about the constraints of the problem. A rule of thumb was coined by Jeff Dean for situations like this:
Ensure your design works if scale changes by 10X or 20X but the right solution for X [is] often not optimal for 100X
https://static.googleusercontent.com/media/research.google.com/en//people/jeff/stanford-295-talk.pdf
So, if in the near future, the system needs to support tens of concurrent users. The simplest that will support tens to hundreds of concurrent users (probably a map or cache with no user sharding is sufficient). That design will have to change before the system can support thousands of concurrent users. Scaling a system is often a good problem to have because it usually indicates a successful project.
I've found different zookeeper definitions across multiple resources. Maybe some of them are taken out of context, but look at them pls:
A canonical example of Zookeeper usage is distributed-memory computation...
ZooKeeper is an open source Apache™ project that provides a centralized infrastructure and services that enable synchronization across a cluster.
Apache ZooKeeper is an open source file application program interface (API) that allows distributed processes in large systems to synchronize with each other so that all clients making requests receive consistent data.
I've worked with Redis and Hazelcast, that would be easier for me to understand Zookeeper by comparing it with them.
Could you please compare Zookeeper with in-memory-data-grids and Redis?
If distributed-memory computation, how does zookeeper differ from in-memory-data-grids?
If synchronization across cluster, than how does it differs from all other in-memory storages? The same in-memory-data-grids also provide cluster-wide locks. Redis also has some kind of transactions.
If it's only about in-memory consistent data, than there are other alternatives. Imdg allow you to achieve the same, don't they?
https://zookeeper.apache.org/doc/current/zookeeperOver.html
By default, Zookeeper replicates all your data to every node and lets clients watch the data for changes. Changes are sent very quickly (within a bounded amount of time) to clients. You can also create "ephemeral nodes", which are deleted within a specified time if a client disconnects. ZooKeeper is highly optimized for reads, while writes are very slow (since they generally are sent to every client as soon as the write takes place). Finally, the maximum size of a "file" (znode) in Zookeeper is 1MB, but typically they'll be single strings.
Taken together, this means that zookeeper is not meant to store for much data, and definitely not a cache. Instead, it's for managing heartbeats/knowing what servers are online, storing/updating configuration, and possibly message passing (though if you have large #s of messages or high throughput demands, something like RabbitMQ will be much better for this task).
Basically, ZooKeeper (and Curator, which is built on it) helps in handling the mechanics of clustering -- heartbeats, distributing updates/configuration, distributed locks, etc.
It's not really comparable to Redis, but for the specific questions...
It doesn't support any computation and for most data sets, won't be able to store the data with any performance.
It's replicated to all nodes in the cluster (there's nothing like Redis clustering where the data can be distributed). All messages are processed atomically in full and are sequenced, so there's no real transactions. It can be USED to implement cluster-wide locks for your services (it's very good at that in fact), and tehre are a lot of locking primitives on the znodes themselves to control which nodes access them.
Sure, but ZooKeeper fills a niche. It's a tool for making a distributed applications play nice with multiple instances, not for storing/sharing large amounts of data. Compared to using an IMDG for this purpose, Zookeeper will be faster, manages heartbeats and synchronization in a predictable way (with a lot of APIs for making this part easy), and has a "push" paradigm instead of "pull" so nodes are notified very quickly of changes.
The quotation from the linked question...
A canonical example of Zookeeper usage is distributed-memory computation
... is, IMO, a bit misleading. You would use it to orchestrate the computation, not provide the data. For example, let's say you had to process rows 1-100 of a table. You might put 10 ZK nodes up, with names like "1-10", "11-20", "21-30", etc. Client applications would be notified of this change automatically by ZK, and the first one would grab "1-10" and set an ephemeral node clients/192.168.77.66/processing/rows_1_10
The next application would see this and go for the next group to process. The actual data to compute would be stored elsewhere (ie Redis, SQL database, etc). If the node failed partway through the computation, another node could see this (after 30-60 seconds) and pick up the job again.
I'd say the canonical example of ZooKeeper is leader election, though. Let's say you have 3 nodes -- one is master and the other 2 are slaves. If the master goes down, a slave node must become the new leader. This type of thing is perfect for ZK.
Consistency Guarantees
ZooKeeper is a high performance, scalable service. Both reads and write operations are designed to be fast, though reads are faster than writes. The reason for this is that in the case of reads, ZooKeeper can serve older data, which in turn is due to ZooKeeper's consistency guarantees:
Sequential Consistency
Updates from a client will be applied in the order that they were sent.
Atomicity
Updates either succeed or fail -- there are no partial results.
Single System Image
A client will see the same view of the service regardless of the server that it connects to.
Reliability
Once an update has been applied, it will persist from that time forward until a client overwrites the update. This guarantee has two corollaries:
If a client gets a successful return code, the update will have been applied. On some failures (communication errors, timeouts, etc) the client will not know if the update has applied or not. We take steps to minimize the failures, but the only guarantee is only present with successful return codes. (This is called the monotonicity condition in Paxos.)
Any updates that are seen by the client, through a read request or successful update, will never be rolled back when recovering from server failures.
Timeliness
The clients view of the system is guaranteed to be up-to-date within a certain time bound. (On the order of tens of seconds.) Either system changes will be seen by a client within this bound, or the client will detect a service outage.
I searched online for awhile about what is "Excessive resource usage" on SQL Azure, still cannot get an idea.
Some articles suggest query takes too long, too much memory etc will cause "Excessive resource usage". But If I use simple query, simple data structure, what will happen?
For example: I get a 1G SQL Azure as session state. Since session is a very small string, and save/delete all the time, I don't think it will grow to 1G for millions of session simultaneously. You can calculate, for 1 million session, 20 char each, only take 20M space, consider 20 minutes expire etc. Cannot even close to 1G. But the queries, should be lots and lots. Each query will be very simple and fast by index.
I wanna know, if this use will be consider as "Excessive resource usage"? Is there any hard number to limit you on the usage?
Btw, as example above, if all happen in same datacenter, so all cost is 1G database which is $10 a month, right?
Unfortunately the answer is 'it depends'. I think that probably the best reference (with guidance) on the SQL Azure Query Throttle is here: TechNet Article on SQL Azure Perormance This will povide details about the metrics that are monitored and the mechanism of the throttle.
The reason that I say it depends is that the throttle is non-deterministic for any given user. This is because the throttle will be activated based on the total load on the node (physical SQL Server in Azure DC). While the subscribers who will get throttled are the subscribers delivering the greatest load the level at which the throttle kicks in will depend on the total load on the node. SO if you are on a quiet node (where other tenant DBs are relatively inactive) then you will be able to put through a bunch more throughput than if you are on a busy node.
It is very appealing to use 1GB SQL Azure DBs for session state storage; you've identified the cost benefits. You are taking a risk though. One way to mitigate this risk is to partition across at least two SQL Azure 1GB DBs and adjust the load yourself based on whether one of the DBs starts hitting the throttle.
Another option if you want determinism for throughput is to use the WIndows Azure Cache to back your sesion state store. The Cache has hard pre-defined limits for query throughput so you can plan for it more easily Azure Caching FAQ including Limits. The Cache approach is probably a bit more expensive but with a lower risk of problems.
What is the difference between Replication and Mirroring in SQL server 2005?
In short, mirroring allows you to have a second server be a "hot" stand-by copy of the main server, ready to take over any moment the main server fails. So mirroring offers fail-over and reliability.
Replication, on the other hand, allows two or more servers to stay "in sync" - that means the secondary servers can answer queries and (depending on setup) actually change data (it will be merged in the sync). You can also use it for local caching, load balancing, etc.
Mirroring is a feature that creates a copy of your database at bit level. Basically you have the same, identical, database in two places. You cannot optionally leave out parts of the database. You can have only one mirror, and the 'mirror' is always offline (it cannot be modified). Mirroring works by shipping the database log as is being created to the mirror and apply (redo-ing) the log on the mirror. Mirroring is a technology for high availability and disaster recoverability.
Replication is a feature that allow 'slices' of a database to be replicated between several sites. The 'slice' can be a set of database objects (ie. tables) but it can also contain parts of a table, like only certain rows (horizontal slicing) or only certain columns to be replicated. You can have multiple replicas and the 'replicas' are available to query and even can be updated. Replication works by tracking/detecting changes (either by triggers or by scanning the log) and shipping the changes, as T-SQL statements, to the subscribers (replicas). Replication is a technology for making data available at off sites and to consolidate data to central sites. Although it is sometimes used for high availability or for disaster recoverability, it is an artificial use for a problem that mirroring and log shipping address better.
There are several types and flavours of replication (merge, transactional, peer-to-peer etc.) and they differ in how they implement change tracking or update propagation, if you want to know more details you should read the MSDN spec on the subject.
Database mirroring is used to increase database uptime and reliability.
Replication is used primarily to distribute portions of your primary database -- the publisher -- to one or more subscriber databases. This is often done to make data available (typically for read only) on remote servers so that remote clients can access the data locally (to them) rather than directly from the publisher across a slower WAN connection. Although, as the previous posts indicate, there are more complex scenarios where updates are permitted on the subscribers. It also can have the benefit of reducing the I/O load on the publisher.