I've got a question about BQ performance in various scenarios, especially revolving around parallelization "under the hood".
I am saving 100M records on a daily basis. At the moment, I am rotating tables every 5 days to avoid high charges due to full table scans.
If I were to run a query with a date range of "last 30 days" (for example), I would be scanning between 6 (if I am at the last day of the partition) and 7 tables.
I could, as an alternative, partition my data into a new table daily. In this case, I will optimize my expenses - as I'm never querying more data than I have too. The question is, will be suffering a performance penalty in terms of getting the results back to the client, because I am now querying potentially 30 or 90 or 365 tables in parallel (Union).
To summarize:
More tables = less data scanned
Less tables =(?) longer response time to the client
Can anyone shed some light on how to find the balance between cost and performance?
A lot depends how you write your queries and how much development costs, but that amount of data doesn't seam like a barrier, and thus you are trying to optimize too early.
When you JOIN tables larger than 8MB, you need to use the EACH modifier, and that query is internally paralleled.
This partitioning means that you can get higher effective read bandwidth because you can read from many of these disks in parallel. Dremel takes advantage of this; when you run a query, it can read your data from thousands of disks at once.
Internally, BigQuery stores tables in
shards; these are discrete chunks of data that can be processed in parallel. If
you have a 100 GB table, it might be stored in 5000 shards, which allows it to be
processed by up to 5000 workers in parallel. You shouldn’t make any assumptions
about the size of number of shards in a table. BigQuery will repartition
data periodically to optimize the storage and query behavior.
Go ahead and create tables for every day, one recommendation is that write your create/patch script that creates tables for far in the future when it runs eg: I create the next 12 months of tables for every day now. This is better than having a script that creates tables each day. And make it part of your deploy/provisioning script.
To read more check out Chapter 11 ■ Managing Data Stored in BigQuery from the book.
Related
We have about 1.7 million products in our eshop, we want to keep record of how many views this products had for 1 year long period, we want to record the views every atleast 2 hours, the question is what structure to use for this task?
Right now we tried keeping stats for 30 days back in records that have 2 columns classified_id,stats where stats is like a stripped json with format date:views,date:views... for example a record would look like
345422,{051216:23212,051217:64233} where 051216,051217=mm/dd/yy and 23212,64233=number of views
This of course is kinda stupid if you want to go 1 year back since if you want to get the sum of views of say 1000 products you need to fetch like 30mb from the database and calculate it your self.
The other way we think of going right now is just to have a massive table with 3 columns classified_id,date,view and store its recording on its own row, this of course will result in a huge table with hundred of millions of rows , for example if we have 1.8 millions of classifieds and keep records 24/7 for one year every 2 hours we need
1800000*365*12=7.884.000.000(billions with a B) rows which while it is way inside the theoritical limit of postgres I imagine the queries on it(say for updating the views), even with the correct indices, will be taking some time.
Any suggestions? I can't even imagine how google analytics stores the stats...
This number is not as high as you think. In current work we store metrics data for websites and total amount of rows we have is much higher. And in previous job I worked with pg database which collected metrics from mobile network and it collected ~2 billions of records per day. So do not be afraid of billions in number of records.
You will definitely need to partition data - most probably by day. With this amount of data you can find indexes quite useless. Depends on planes you will see in EXPLAIN command output. For example that telco app did not use any indexes at all because they would just slow down whole engine.
Another question is how quick responses for queries you will need. And which steps in granularity (sums over hours/days/weeks etc) for queries you will allow for users. You may even need to make some aggregations for granularities like week or month or quarter.
Addition:
Those ~2billions of records per day in that telco app took ~290GB per day. And it meant inserts of ~23000 records per second using bulk inserts with COPY command. Every bulk was several thousands of records. Raw data were partitioned by minutes. To avoid disk waits db had 4 tablespaces on 4 different disks/ arrays and partitions were distributed over them. PostreSQL was able to handle it all without any problems. So you should think about proper HW configuration too.
Good idea also is to move pg_xlog directory to separate disk or array. No just different filesystem. It all must be separate HW. SSDs I can recommend only in arrays with proper error check. Lately we had problems with corrupted database on single SSD.
First, do not use the database for recording statistics. Or, at the very least, use a different database. The write overhead of the logs will degrade the responsiveness of your webapp. And your daily backups will take much longer because of big tables that do not need to be backed up so frequently.
The "do it yourself" solution of my choice would be to write asynchronously to log files and then process these files afterwards to construct the statistics in your analytics database. There is good code snippet of async write in this response. Or you can benchmark any of the many loggers available for Java.
Also note that there are products like Apache Kafka specifically designed to collect this kind of information.
Another possibility is to create a time series in column oriented database like HBase or Cassandra. In this case you'd have one row per product and as many columns as hits.
Last, if you are going to do it with the database, as #JosMac pointed, create partitions, avoid indexes as much as you can. Set fillfactor storage parameter to 100. You can also consider UNLOGGED tables. But read thoroughly PostgreSQL documentation before turning off the write-ahead log.
Just to raise another non-RDBMS option for you (so a little off topic), you could send text files (CSV, TSV, JSON, Parquet, ORC) to Amazon S3 and use AWS Athena to query it directly using SQL.
Since it will query free text files, you may be able to just send it unfiltered weblogs, and query them through JDBC.
Inserting data to U-SQL table is taking too much time. We are using partitioned tables to recalculate previously processed data. Insertion for the first time took almost 10-12 minutes on three tables with 11, 5 and 1 partitions and parallelism was set to 10. Second time insertion of same data took almost 4 hours. Currently we are using year based partitions. We tested insertion and querying without adding partitions and performance was much better. Is this an issue with partitioned tables?
It is very strange that the same job would be taking that much longer for the same data and script executed with the same degree of parallelism. If you look at the job graph (or the vertex execution information) from within VisualStudio, can you see where the time was being spent?
Note that (coarse-grained) partitions are more of a data life-cycle management feature that allows you to address individual partitions of a table, and not necessarily a performance feature (although partition elimination can help with query performance). But it should not go from minutes to hours with the same script, resources and data.
I'm using BigQuery for ~5 billion rows that can be partitioned on ~1 million keys.
Since our queries are usually by partition key, is it possible to create ~1 million tables (1 table / key) to limit the total number of bytes processed?
We also need to query all of the data together at times, which is easy to do by putting it all in one table, but I'm hoping to use the same platform for partitioned analysis as bulk analytics.
That might work, but partitioning your table this finely is highly discouraged. You might be better off partitioning your data into a smaller number of tables, say 10 or 100, and querying just the one(s) you need.
What do I mean by discouraged? First, each of those million tables will get charged a minimum of 10 MB for storage. So you'll get charged for 9 TB of storage, when you likely have a lot less data than that. Second, you'll likely hit rate limits when you try to create that many tables. Third, managing a million tables is very tricky; the BigQuery UI will likely not be much help. Fourth, you'll make engineers on the BigQuery exceedingly grumpy, and they'll start trying to figure out whether we need to raise the minimum size for tables.
Also, if you do want to sometimes query all of your data, partitioning this finely is likely going to make things difficult for you, unless you are willing to store your data multiple times. You can only reference 1000 tables in a query, and each one you reference causes you to take a performance hit.
Folks, I'm using BigQuery as a superfast database for my analytics queries, but I'm very disappointed with its performance.
Let me show you the numbers:
Just one Table at "from" clause
Select about 15 fields with group by each, about 5 fields with SUM()
Total table rows: 3.7 millions
Total rows returned: 830K
When I execute this query on BigQuery's console, it takes about 1 minute to process. Is this ok for you? I was expecting that it will return in about 2 seconds... If I execute this query on a columnar database, like Sybase IQ, it takes less than 2 seconds.
Big Query is a highly scalable database, before being a "super fast" database. It's designed to process HUGE amount of data distributing the processing among several different machines using a technique named Dremel. Because it's designed to use several machines and parallel processing, you should expect to have super-scalability with a good performance.
For example: analyzing all the wikipedia revisions in 5-10 seconds isn't bad, is it? But even a much smaller table would take about the same time.
Sybase IQ is often installed in a single database and it doesn't use Dremel. That said, it's going to be faster than Big Query in many scenarios...as designed.
Cheers!
Since you are returning 830k rows and BQ is always creating a temporary result table, the creation is more than a small result.
Have you turned on large results?
We are working in a shared environment and sometime loads ( table creation ) takes a while.
Certainly the performance differ from a dedicated environment. You get your dedicated environment for 20K$ a month.
I have a database table with about 700 millions rows plus (growing exponentially) of time based data.
Fields:
PK.ID,
PK.TimeStamp,
Value
I also have 3 other tables grouping this data into Days, Months, Years which contains the sum of the value for each ID in that time period. These tables are updated nightly by a SQL job, the situation has arisen where by the tables will need to updated on the fly when the data in the base table is updated, this can be however up to 2.5 million rows at a time (not very often, typically around 200-500k up to every 5 minutes), is this possible without causing massive performance hits or what would be the best method for achieving this?
N.B
The daily, monthly, year tables can be changed if needed, they are used to speed up queries such as 'Get the monthly totals for these 5 ids for the last 5 years', in raw data this is about 13 million rows of data, from the monthly table its 300 rows.
I do have SSIS available to me.
I cant afford to lock any tables during the process.
700M recors in 5 months mean 8.4B in 5 years (assuming data inflow doesn't grow).
Welcome to the world of big data. It's exciting here and we welcome more and more new residents every day :)
I'll describe three incremental steps that you can take. The first two are just temporary - at some point you'll have too much data and will have to move on. However, each one takes more work and/or more money so it makes sense to take it a step at a time.
Step 1: Better Hardware - Scale up
Faster disks, RAID, and much more RAM will take you some of the way. Scaling up, as this is called, breaks down eventually, but if you data is growing linearly and not exponentially, then it'll keep you floating for a while.
You can also use SQL Server replication to create a copy of your database on another server. Replication works by reading transaction logs and sending them to your replica. Then you can run the scripts that create your aggregate (daily, monthly, annual) tables on a secondary server that won't kill the performance of your primary one.
Step 2: OLAP
Since you have SSIS at your disposal, start discussing multidimensional data. With good design, OLAP Cubes will take you a long way. They may even be enough to manage billions of records and you'll be able to stop there for several years (been there done that, and it carried us for two years or so).
Step 3: Scale Out
Handle more data by distributing the data and its processing over multiple machines. When done right this allows you to scale almost linearly - have more data then add more machines to keep processing time constant.
If you have the $$$, use solutions from Vertica or Greenplum (there may be other options, these are the ones that I'm familiar with).
If you prefer open source / byo, use Hadoop, log event data to files, use MapReduce to process them, store results to HBase or Hypertable. There are many different configurations and solutions here - the whole field is still in its infancy.
Indexed views.
Indexed views will allow you to store and index aggregated data. One of the most useful aspects of them is that you don't even need to directly reference the view in any of your queries. If someone queries an aggregate that's in the view, the query engine will pull data from the view instead of checking the underlying table.
You will pay some overhead to update the view as data changes, but from your scenario it sounds like this would be acceptable.
Why don't you create monthly tables, just to save the info you need for that months. It'd be like simulating multidimensional tables. Or, if you have access to multidimensional systems (oracle, db2 or so), just work with multidimensionality. That works fine with time period problems like yours. At this moment I don't have enough info to give you, but you can learn a lot about it just googling.
Just as an idea.