Best Big Data Database for time series - database

i'm looking for the best database for my big data project.
We are collecting data from some sensors. Every row has about one hundred column.
every day we store some milions of rows.
The most common query is for retreiving data for one sensor in a range of date.
at the moment i use a percona mysql cluster. when i ask data for a range on some days, the response is fast. The problem is when i ask data for a month.
The database is perfectly optimized, but the response time is not acceptable.
I would like to change percona cluster with a database able to perform query in parallel on all the nodes to improve response time.
With Cassandra i could partition data accross nodes (maybe based on the current date) but i have read that cassandra cannot read data between partition in parallel, but i have to create a query for every day. (i don't know why)
Is there a database that manage shard queries automatically, so i can distribute data across all nodes?

With Cassandra, If you split your data across multiple partitions, you still can read data between partition in parallel by executing multiples queries asynchronously.
Cassandra drivers help you handle this, see execute_concurrent from the python driver.
Moreover, the cassandra driver is aware of the data partitioning, it knows which node holds which data. So when reading or writing, it chooses an appropriate node to send the query, according to the driver load balancing policy (specifically with the TokenAwarePolicy).
Thus, the client acts as a load balancer, and your request is processed in parallel by the available nodes.

Related

Best practices for keeping a custom table up-to-date from data derived from other Snowflake DB's in our warehouse

(Submitting on behalf of a Snowflake User)
We have a database that stores raw data from all our local sources. My team has it's own environment in which we have full permissions to create standardized feeds and/or tables/views etc that is ready to consume through Power BI. A few additional details:
The final 'feed' tables are derived through SQL statements and most are pulling from more than one table of our 'raw' data.
Raw table data is updated daily.
My question is what is the best operation to keep the tables fully updated and what is the standard work flow for this operations? Our current understanding is one of these processes is the best:
Using COPY INTO <stage> then COPY INTO <table>.
Using STREAMS to add incremental data.
Using PIPES (may be the same as STREAMS)
Or simplify our feeds to one table sources and use a materialized view.
We'd ideally like to avoid views to improve consumption speed at the power bi level.
Tasks have been recommended as it seems like a good fit since they only need to update the final table once per day.
(https://docs.snowflake.net/manuals/sql-reference/sql/create-task.html)
Any other recommendations??? THANKS!
We have a similar scenario where we have our raw datalake tables being updated real time from files in S3. Those raw tables are loaded via snowpipe using the auto ingest feature.
In turn, we have a data mart which contains facts about the raw data. To update the data mart, we have created streams on top of the raw tables to track changes. We then use tasks run at a given frequency (every five minutes in our case) to update the data mart from the changed data in the raw tables. Using streams allows us to limit processing to only changed data, without having to track last update dates, etc.

Can I use Hadoop to speed up a slow SQL stored procedure?

The problem:
I have 2 SQL Server databases from 2 different applications. They describe different aspects of industrial machines: one is about "how many consumables were spent per order", the other is about "how many good/bad production items were produced per operator". Sometimes many operators are working on 1 order one after another, sometimes one operator is working on multiple small orders, and there is no connection Order-Operator in the database.
I want to have united fact table, where for every timestamp I know MachineID, OrderID and OperatorID. If a timestamp exists in DB1, then the record will have numeric measures from it (Consumables); if it exists in DB2, then it will have numeric measures from DB2 (good/bad production items). If it exists in both databases, then it have all numeric measures. A simple UNION ALL is not enough, because I want to have MachineID, OrderID and OperatorID for every record.
I created a T-SQL stored procedure to make FULL JOIN by timestamp and MachineID. But on large data sets (multiple machines, multiple customers) it becomes very slow. Both applications support editing history, so I need to merge full history from both databases at every nightly load.
To speed up the process, I would like to put calculations into multiple parallel threads, separated by Customer, MachineID, and Year.
I tried to do it by using SQL Server stored procedures, running in parallel by SQL Agent with different parameters, but I found that it didn't help the performance. Instead it created multiple deadlocks when updating staging and final tables.
I am looking for an alternative way to resolve this problem, but I don't know what is the right tool. Can Hadoop or similar parallel processing tool help with this task?
I am looking for solution with minimal cost, because it is needed for just one specific task. For everything else, SQL Server and PowerBI reporting are working just fine for me.
Hadoop seems hard to justify in this use case, given limited scope. The thing about Hadoop is that it scales well not only due to parallel processing but thanks to parallel IO, when data is distributed across multiple servers/storage media. Unless you happy to copy all data to HDFS distributed among multiple nodes, it likely will not help much. If you want to spin up a Hadoop cluster and run multiple jobs querying single SQL server, it'll likely end up badly for the later.
Have you considered optimizations which will allow you to limit the amount of data you processing nightly?
E.g. what is 'timestamp' field? Does it reflect last update time? Can you use it to filter rows which haven't been updated since the previous run?
Even if the 'timestamp' is not the time of last updates, can you add an "updateTime" field and triggers on updates which will populate the field, so you don't need to import rows which have not changed since the previous run? If you build an index on the field, then, if the number of updates during the day is not high relative to total table size, a query with a filter on such field will hit the index, and fetching of incremental changes should be fast.
Another thing to consider - are those DBs running on the same node/SQL server? Access to remote DBs is slow, so if that's the case, think about how to fix this first.

How to administrate storage of ClickHouse server in a Cluster when disks get full

I'm setting up a ClickHouse server in cluster, but one of the things that doesn't appear in the documentation is how to manage very large amount of data, it says that it can handle up to petabytes of data, but you can't store that much data in single server. You usually will have a few teras in each.
So my question is, how can I handle it to store in a node of the cluster, and then when it requires more space, add another, will it handle the distribution to the new server automatically or will I have to play with the weights in the shard distribution.
When you have more than 1 disk in one server, how can it use them all to store the data?
Is there a way to store very old data in the cloud and download it if needed? For example all data older than 2 years can be stored in Amazon S3 as it will be hardly requested and in case it is, it will take a longer time to retreive the data but wouldn't be a problem.
What solution would you find to this? Handling an ever exapanding database to avoid disk space issues in the future.
Thanks
I will assume that you use standard configuration for the ClickHouse cluster: several shards consisting of 2-3 replica nodes, and on each of these nodes a ReplicatedMergeTree table containing data for its respective shard. There are also Distributed tables created on one or more nodes that are configured to query the nodes of the cluster (relevant section in the docs).
When you add a new shard, old data is not moved to it automatically. Recommended approach is indeed to "play with the weights" as you have put it, i.e. increase the weight of the new node until the volume of data is even. But if you want to rebalance the data immediately, you can use the ALTER TABLE RESHARD command. Read the docs carefully and keep in mind various limitations of this command, e.g. it is not atomic.
When you have more than 1 disk in one server, how can it use them all to store the data?
Please read the section on configuring RAID in the administration tips.
Is there a way to store very old data in the cloud and download it if needed? For example all data older than 2 years can be stored in Amazon S3 as it will be hardly requested and in case it is, it will take a longer time to retreive the data but wouldn't be a problem.
MergeTree tables in ClickHouse are partitioned by month. You can use ALTER TABLE DETACH/ATTACH PARTITION commands to manipulate partitions. You can e.g. at the start of each month detach the partition for some older month and back it up to Amazon S3. Or you can setup a cluster of cheaper machines with ample disk space and manually move old partitions there. If your queries always include a filter on date, irrelevant partitions will be skipped automatically, else you can setup two Distributed tables: table_recent and table_all (with the cluster config including the nodes with old partitions).
Version 19.15 introduced multidisk strorage configuration. 20.1 introduces time-based data rearrangements.

Efficiently synchronize streaming input with large SQL database

I've got an efficiency problem on my hands and I'm looking for ways to solve it. Here is the situation:
I have streaming row-based data coming into a system from an online source, each with unique id.
I have an SQL database with existing row-based data, indexed by id.
I need to update the SQL database with new data streaming in if it differs from the existing data already in the database.
The obvious solution is:
Read incoming row
Read corresponding row in database
If data differs, update the database
The large number of round-trips is making this algorithm infeasibly slow.
The alternative solution I have is to read the entire SQL database into memory, and compare with new incoming data. This eliminates the round-trips but the memory required makes this also infeasible.
So, what alternatives do I have?
Split the database into multiple parts. (To address the size issue)
Design an algorithm to cache some of these database pieces into memory. (Speed)
Based on the incoming id, hash the query to the relevant database.
If step 2 is not possible to be achieved efficiently, this wont work.

A huge data storage problem

I'm starting to design a new application that will be used by about 50000 devices. Each device generates about 1440 registries a day, this means that will be stored over 72 million of registries per day. These registries keep coming every minute, and I must be able to query this data by a Java application (J2EE). So it need to be fast to write, fast to read and indexed to allow report generation.
Devices only insert data and the J2EE application will need to read then occasionally.
Now I'm looking to software alternatives to support this kind of operation.
Putting this data on a single table would lead to a catastrophic condition, because I won't be able to use this data due to its amount of data stored over a year.
I'm using Postgres, and database partitioning seems not to be a answer, since I'd need to partition tables by month, or may be more granular approach, days for example.
I was thinking on a solution using SQLite. Each device would have its own SQLite database, than the information would be granular enough for good maintenance and fast insertions and queries.
What do you think?
Record only changes of device positions - most of the time any device will not move - a car will be parked, a person will sit or sleep, a phone will be on unmoving person or charged etc. - this would make you an order of magnitude less data to store.
You'll be generating at most about 1TB a year (even when not implementing point 1), which is not a very big amount of data. This means about 30MB/s of data, which single SATA drive can handle.
Even a simple unpartitioned Postgres database on not too big hardware should manage to handle this. The only problem could be when you'll need to query or backup - this can be resolved by using a Hot Standby mirror using Streaming Replication - this is a new feature in soon to be released PostgreSQL 9.0. Just query against / backup a mirror - if it is busy it will temporarily and automatically queue changes, and catch up later.
When you really need to partition do it for example on device_id modulo 256 instead of time. This way you'd have writes spread out on every partition. If you partition on time just one partition will be very busy on any moment and others will be idle. Postgres supports partitioning this way very well. You can then also spread load to several storage devices using tablespaces, which are also well supported in Postgres.
Time-interval partitioning is a very good solution, even if you have to roll your own. Maintaining separate connections to 50,000 SQLite databases is much less practical than a single Postgres database, even for millions of inserts a day.
Depending on the kind of queries that you need to run against your dataset, you might consider partitioning your remote devices across several servers, and then query those servers to write aggregate data to a backend server.
The key to high-volume tables is: minimize the amount of data you write and the number of indexes that have to be updated; don't do UPDATEs or DELETEs, only INSERTS (and use partitioning for data that you will delete in the future—DROP TABLE is much faster than DELETE FROM TABLE!).
Table design and query optimization becomes very database-specific as you start to challenge the database engine. Consider hiring a Postgres expert to at least consult on your design.
Maybe it is time for a db that you can shard over many machines? Cassandra? Redis? Don't limit yourself to sql db's.
Database partition management can be automated; time-based partitioning of the data is a standard way of dealihg with this type of problem, and I'm not sure that I can see any reason why this can't be done with PostgreSQL.
You have approximately 72m rows per day - assuming a device ID, datestamp and two floats for coordinates you will have (say) 16-20 bytes per row plus some minor page metadata overhead. A back-of-fag-packet capacity plan suggests around 1-1.5GB of data per day, or 400-500GB per year, plus indexes if necessary.
If you can live with periodically refreshed data (i.e. not completely up to date) you could build a separate reporting table and periodically update this with an ETL process. If this table is stored on separate physical disk volumes it can be queried without significantly affecting the performance of your transactional data.
A separate reporting database for historical data would also allow you to prune your operational table by dropping older partitions, which would probably help with application performance. You could also index the reporting tables and create summary tables to optimise reporting performance.
If you need low latency data (i.e. reporting on up-to-date data), it may also be possible to build a view where the lead partitions are reported off the operational system and the historical data is reported from the data mart. This would allow the bulk queries to take place on reporting tables optimised for this, while relatively small volumes of current data can be read directly from the operational system.
Most low-latency reporting systems use some variation of this approach - a leading partition can be updated by a real-time process (perhaps triggers) and contains relatively little data, so it can be queried quickly, but contains no baggage that slows down the update. The rest of the historical data can be heavily indexed for reporting. Partitioning by date means that the system will automatically start populating the next partition, and a periodic process can move, re-index or do whatever needs to be done for the historical data to optimise it for reporting.
Note: If your budget runs to PostgreSQL rather than Oracle, you will probably find that direct-attach storage is appreciably faster than a SAN unless you want to spend a lot of money on SAN hardware.
That is a bit of a vague question you are asking. And I think you are not facing a choice of database software, but an architectural problem.
Some considerations:
How reliable are the devices, and how
well are they connected to the
querying software?
How failsafe do
you need the storage to be?
How much extra processing power do the devices
have to process your queries?
Basically, your idea of a spatial partitioning is a good idea. That does not exclude a temporal partition, if necessary. Whether you do that in postgres or sqlite depends on other factors, like the processing power and available libraries.
Another consideration would be whether your devices are reliable and powerful enough to handle your queries. Otherwise, you might want to work with a centralized cluster of databases instead, which you can still query in parallel.

Resources