I'm trying to think of an algorithm to solve this problem I have. It's not a HW problem, but for a side project I'm working on.
There's a table A that has about (order of) 10^5 rows and adds new in the order of 10^2 every day.
Table B has on the order of 10^6 rows and adds new at 10^3 every day. There's a one to many relation from A to B (many B rows for some row in A).
I was wondering how I could do continuous aggregates for this kind of data. I would like to have a job that runs every ~10mins and does this: For every row in A, find every row in B related to it that were created in the last day, week and month (and then sort by count) and save them in a different DB or cache them.
If this is confusing, here's a practical example: Say table A has Amazon products and table B has product reviews. We would like to show a sorted list of products with highest reviews in the last 4hrs, day, week etc. New products and reviews are added at a fast pace, and we'd like the said list to be as up-to-date as possible.
Current implementation I have is just a for loop (pseudo-code):
result = []
for product in db_products:
reviews = db_reviews(product_id=product.id, create>=some_time)
reviews_count = len(reviews)
result[product]['reviews'] = reviews
result[product]['reviews_count'] = reviews_count
sort(result, by=reviews_count)
return result
I do this every hour, and save the result in a json file to serve. The problem is that this doesn't really scale well, and takes a long time to compute.
So, where could I look to solve this problem?
UPDATE:
Thank you for your answers. But I ended up learning and using Apache Storm.
Summary of requirements
Having two bigger tables in a database, you need regularly creating some aggregates for past time periods (hour, day, week etc.) and store the results in another database.
I will assume, that once a time period is past, there are no changes to related records, in other words, the aggregate for past period has always the same result.
Proposed solution: Luigi
Luigi is framework for plumbing dependent tasks and one of typical uses is calculating aggregates for past periods.
The concept is as follows:
write simple Task instance, which defines required input data, output data (called Target) and process to create the target output.
Tasks can be parametrized, typical parameter is time period (specific day, hour, week etc.)
Luigi can stop tasks in the middle and start later. It will consider any task, for which is target already existing to be completed and will not rerun it (you would have to delete the target content to let it rerun).
In short: if the target exists, the task is done.
This works for multiple types of targets like files in local file system, on hadoop, at AWS S3, and also in database.
To prevent half done results, target implementations take care of atomicity, so e.g. files are first created in temporary location and are moved to final destination just after they are completed.
In databases there are structures to denote, that some database import is completed.
You are free to create your own target implementations (it has to create something and provide method exists to check, the result exists.
Using Luigi for your task
For the task you describe you will probably find everything you need already present. Just few tips:
class luigi.postgres.CopyToTable allowing to store records into Postgres database. The target will automatically create so called "marker table" where it will mark all completed tasks.
There are similar classes for other types of databases, one of them using SqlAlchemy which shall probably cover the database you use, see class luigi.contrib.sqla.CopyToTable
At Luigi doc is working example of importing data into sqlite database
Complete implementation is beyond extend feasible in StackOverflow answer, but I am sure, you will experience following:
The code to do the task is really clear - no boilerplate coding, just write only what has to be done.
nice support for working with time periods - even from command line, see e.g. Efficiently triggering recurring tasks. It even takes care of not going too far in past, to prevent generating too many tasks possibly overloading your servers (default values are very reasonably set and can be changed).
Option to run the task on multiple servers (using central scheduler, which is provided with Luigi implementation).
I have processed huge amounts of XML files with Luigi and also made some tasks, importing aggregated data into database and can recommend it (I am not author of Luigi, I am just happy user).
Speeding up database operations (queries)
If your task suffers from too long execution time to perform the database query, you have few options:
if you are counting reviews per product by Python, consider trying SQL query - it is often much faster. It shall be possible to create SQL query which uses count on proper records and returns directly the number you need. With group by you shall even get summary information for all products in one run.
set up proper index, probably on "reviews" table on "product" and "time period" column. This shall speed up the query, but make sure, it does not slow down inserting new records too much (too many indexes can cause that).
It might happen, that with optimized SQL query you will get working solution even without using Luigi.
Data Warehousing? Summary tables are the right way to go.
Does the data change (once it is written)? If it does, then incrementally updating Summary Tables becomes a challenge. Most DW applications do not have that problem
Update the summary table (day + dimension(s) + count(s) + sum(s)) as you insert into the raw data table(s). Since you are getting only one insert per minute, INSERT INTO SummaryTable ... ON DUPLICATE KEY UPDATE ... would be quite adequate, and simpler than running a script every 10 minutes.
Do any reporting from a summary table, not the raw data (the Fact table). It will be a lot faster.
My Blog on Summary Tables discusses details. (It is aimed at bigger DW applications, but should be useful reading.)
I agree with Rick, summary tables make the most sense for you. Update the summary tables every 10 minutes and just pull data from it, as user's request summaries.
Also, make sure that your DB is indexed properly for performance. I'm sure db_products.id set as a unique index. but, also make sure that db_products.create is defined as a DATE or DATETIME and also indexed since you are using it in your WHERE statement.
Related
Background:
I have multiple asset tables stored in a redshift database for each city, 8 cities in total. These asset tables display status updates on an hourly basis. 8 SQL tables and about 500 mil rows of data in a year.
(I also have access to the server that updates this data every minute.)
Example: One market can have 20k assets displaying 480k (20k*24 hrs) status updates a day.
These status updates are in a raw format and need to undergo a transformation process that is currently written in a SQL view. The end state is going into our BI tool (Tableau) for external stakeholders to look at.
Problem:
The current way the data is processed is slow and inefficient, and probably not realistic to run this job on an hourly basis in Tableau. The status transformation requires that I look back at 30 days of data, so I do need to look back at the history throughout the query.
Possible Solutions:
Here are some solutions that I think might work, I would like to get feedback on what makes the most sense in my situation.
Run a python script that looks at the most recent update and query the large history table 30 days as a cron job and send the result to a table in the redshift database.
Materialize the SQL view and run an incremental refresh every hour
Put the view in Tableau as a datasource and run an incremental refresh every hour
Please let me know how you would approach this problem. My knowledge is in SQL, limited Data Engineering experience, Tableau (Prep & Desktop) and scripting in Python or R.
So first things first - you say that the data processing is "slow and inefficient" and ask how to efficiently query a large database. First I'd look at how to improve this process. You indicate that the process is based on the past 30 days of data - is the large tables time sorted, vacuumed and analyzed? It is important to take maximum advantage of metadata when working with large tables. Make sure your where clauses are effective at eliminating fact table block - don't rely on dimension table where clauses to select the date range.
Next look at your distribution keys and how these are impacting the need for your critical query to move large amounts of data across the network. The internode network has the lowest bandwidth in a Redshift cluster and needlessly pushing lots of data across it will make things slow and inefficient. Using EVEN distribution can be a performance killer depending on your query pattern.
Now let me get to your question and let me paraphrase - "is it better to use summary tables, materialized views, or external storage (tableau datasource) to store summary data updated hourly?" All 3 work and each has its own pros and cons.
Summary tables are good because you can select the distribution of the data storage and if this data needs to be combined with other database tables it can be done most efficiently. However, there is more data management to be performed to keep this data up to data and in sync.
Materialized views are nice as there is a lot less management action to worry about - when the data changes, just refresh the view. The data is still in the database so is is easy to combine with other data tables but since you don't have control over storage of the data these action may not be the most efficient.
External storage is good in that the data is in your BI tool so if you need to refetch the results during the hour the data is local. However, it is not locked into your BI tool and far less efficient to combine with other database tables.
Summary data usually isn't that large so how it is stored isn't a huge concern and I'm a bit lazy so I'd go with a materialized view. Like I said at the beginning I'd first look at the "slow and inefficient" queries I'm running every hour first.
Hope this helps
We are migrating some data from our production database and would like to archive most of this data in the Cloud Datastore.
Eventually we would move all our data there, however initially focusing on the archived data as a test.
Our language of choice is Python, and have been able to transfer data from mysql to the datastore row by row.
We have approximately 120 million rows to transfer and at a one row at a time method will take a very long time.
Has anyone found some documentation or examples on how to bulk insert data into cloud datastore using python?
Any comments, suggestions is appreciated thank you in advanced.
There is no "bulk-loading" feature for Cloud Datastore that I know of today, so if you're expecting something like "upload a file with all your data and it'll appear in Datastore", I don't think you'll find anything.
You could always write a quick script using a local queue that parallelizes the work.
The basic gist would be:
Queuing script pulls data out of your MySQL instance and puts it on a queue.
(Many) Workers pull from this queue, and try to write the item to Datastore.
On failure, push the item back on the queue.
Datastore is massively parallelizable, so if you can write a script that will send off thousands of writes per second, it should work just fine. Further, your big bottleneck here will be network IO (after you send a request, you have to wait a bit to get a response), so lots of threads should get a pretty good overall write rate. However, it'll be up to you to make sure you split the work up appropriately among those threads.
Now, that said, you should investigate whether Cloud Datastore is the right fit for your data and durability/availability needs. If you're taking 120m rows and loading it into Cloud Datastore for key-value style querying (aka, you have a key and an unindexed value property which is just JSON data), then this might make sense, but loading your data will cost you ~$70 in this case (120m * $0.06/100k).
If you have properties (which will be indexed by default), this cost goes up substantially.
The cost of operations is $0.06 per 100k, but a single "write" may contain several "operations". For example, let's assume you have 120m rows in a table that has 5 columns (which equates to one Kind with 5 properties).
A single "new entity write" is equivalent to:
+ 2 (1 x 2 write ops fixed cost per new entity)
+ 10 (5 x 2 write ops per indexed property)
= 12 "operations" per entity.
So your actual cost to load this data is:
120m entities * 12 ops/entity * ($0.06/100k ops) = $864.00
I believe what you are looking for is the put_multi() method.
From the docs, you can use put_multi() to batch multiple put operations. This will result in a single RPC for the batch rather than one for each of the entities.
Example:
# a list of many entities
user_entities = [ UserEntity(name='user %s' % i) for i in xrange(10000)]
users_keys = ndb.put_multi(user_entities) # keys are in same order as user_entities
Also to note, from the docs is that:
Note: The ndb library automatically batches most calls to Cloud Datastore, so in most cases you don't need to use the explicit batching operations shown below.
That said, you may still, as suggested by , use a task queue (I prefer the deferred library) in order to batch-put a lot of data in the background.
As an update to the answer of #JJ Geewax, as of July 1st, 2016
the cost of read and write operations have changed as explained here: https://cloud.google.com/blog/products/gcp/google-cloud-datastore-simplifies-pricing-cuts-cost-dramatically-for-most-use-cases
So writing should have gotten cheaper for the described case, as
writing a single entity only costs 1 write regardless of indexes and will now cost $0.18 per 100,000
I'm working on a project that allows users to enter SQL queries with parameters, that SQL query will be executed over a period of time they decide (say every 2 hours for 6 months) and then get the results back to their email address.
They'll get it in the form of an HTML-email message, so what the system basically does is run the queries, and generate HTML that is then sent to the user.
I also want to save those results, so that a user can go on our website and look at previous results.
My question is - what data do I save?
Do I save the SQL query with those parameters (i.e the date parameters, so he can see the results relevant to that specific date). This means that when the user clicks on this specific result, I need to execute the query again.
Save the HTML that was generated back then, and simply display it when the user wishes to see this result?
I'd appreciate it if somebody would explain the pros and cons of each solution, and which one is considered the best & the most efficient.
The archive will probably be 1-2 months old, and I can't really predict the amount of rows each query will return.
Thanks!
Specifically regarding retrieving the results from queries that have been run previously I would suggest saving the results to be able to view later rather than running the queries again and again. The main benefits of this approach are:
You save unnecessary computational work re-running the same queries;
You guarantee that the result set will be the same as the original report. For example if you save just the SQL then the records queried may have changed since the query was last run or records may have been added / deleted.
The disadvantage of this approach is that it will probably use more disk space, but this is unlikely to be an issue unless you have queries returning millions of rows (in which case html is probably not such a good idea anyway).
If I would create such type of application then
I will have some common queries like get by current date,current time , date ranges, time ranges, n others based on my application for the user to select easily.
Some autocompletions for common keywords.
If the data gets changed frequently there is no use saving html, generating new one is good option
The crucial difference is that if data changes, new query will return different result than what was saved some time ago, so you have to decide if the user should get the up to date data or a snapshot of what the data used to be.
If relevant data does not change, it's a matter of whether the queries will be expensive, how many users will run them and how often, then you may decide to save them instead of re-running queries, to improve performance.
I am working on a web application for downloading resources of an unimportant type. It's written in python using the flask web framework. I use the SQLAlchemy DB system.
It has a user authentication system and you can download the resources only while logged in.
What I am trying to do is a download history chart for every resource and every user. To elaborate, each user could see two charts of their download activity on their profile page, for the last 7 days and the last year respectively. Each resource would also have a similar pair of charts, but they would instead visualize how many times the resource itself was downloaded in the time periods.
Here is an example screenshot of the charts
(Don't have enough reputation to embed images)
http://dl.dropbox.com/u/5011799/Selection_049.png
The problem is, I can't seem to figure out what the best way to store the downloads in a database would be. I found 2 ways that are relatively easy to implement and should work:
1) I could store the download count for each day in the last week in separate fields and every 24 hours just get rid of the first one and move them to the left by 1. This, however, seems like a kind of a hacky way to do this.
2) I could also create a separate table for the downloads and every time a user downloads a resource I would insert a row into the table with the Datetime, user_id of the downloader and the resource_id of the downloaded resource. This would allow me to do some nice querying of time periods etc. The problem with that configuration could be the row count in the table. I have no idea how heavily the website is going to be used, but if I do the math with 1000 downloads / day, I am going to end up with over 360k rows in just the first year. I don't know how fast that would to perform. I know I could just archive old entries if performace started being a huge problem.
I would like to know whether the 2nd option would be fast enough for a web app and what configuration you would use.
Thanks in advance.
I recommend the second approach, with periodic aggregation to improve performance.
Storing counts by day will force you to SELECT the existing count so that you can either add to it with an UPDATE statement or know that you need to INSERT a new record. That's two trips to the database on every download. And if things get out of whack, there's really no easy way to determine what happened or what the correct numbers ought to be. (You're not saving information about the individual events.) That's probably not a significant concern for a simple download count, but if this were sensitive information it might matter.
The second approach simply requires a single INSERT for each download, and because each event is stored separately, it's easy to troubleshoot. And, as you point out, you can slice this data any way you like.
As for performance, 360,000 rows is trivial for a modern RDBMS on contemporary hardware, but you do want to make sure you have an index on date, username/resource name or any other columns that will be used to select data.
Still, you might have more volume than you expect, or maybe your DB is iffy (I'm not familiar with SQLAlchemy). To reduce your row count you could create a weekly batch process (yeah, I know, batch ain't dead despite what some people say) during non-peak hours to create summary records by week.
It would probably be easiest to create your summary records in a different table that is simply keyed by week and year, or start/end dates, depending on how you want to use it. After you've generated the weekly summary for a period, you can archive or delete the daily detail records for that period.
I'm making an app that has a need for reverse searches. By this, I mean that users of the app will enter search parameters and save them; then, when any new objects get entered onto the system, if they match the existing search parameters that a user has saved, a notification will be sent, etc.
I am having a hard time finding solutions for this type of problem.
I am using Django and thinking of building the searches and pickling them using Q objects as outlined here: http://www.djangozen.com/blog/the-power-of-q
The way I see it, when a new object is entered into the database, I will have to load every single saved query from the db and somehow run it against this one new object to see if it would match that search query... This doesn't seem ideal - has anyone tackled such a problem before?
At the database level, many databases offer 'triggers'.
Another approach is to have timed jobs that periodically fetch all items from the database that have a last-modified date since the last run; then these get filtered and alerts issued. You can perhaps put some of the filtering into the query statement in the database. However, this is a bit trickier if notifications need to be sent if items get deleted.
You can also put triggers manually into the code that submits data to the database, which is perhaps more flexible and certainly doesn't rely on specific features of the database.
A nice way for the triggers and the alerts to communicate is through message queues - queues such as RabbitMQ and other AMQP implementations will scale with your site.
The amount of effort you use to solve this problem is directly related to the number of stored queries you are dealing with.
Over 20 years ago we handled stored queries by treating them as minidocs and indexing them based on all of the must have and may have terms. A new doc's term list was used as a sort of query against this "database of queries" and that built a list of possibly interesting searches to run, and then only those searches were run against the new docs. This may sound convoluted, but when there are more than a few stored queries (say anywhere from 10,000 to 1,000,000 or more) and you have a complex query language that supports a hybrid of Boolean and similarity-based searching, it substantially reduced the number we had to execute as full-on queries -- often no more that 10 or 15 queries.
One thing that helped was that we were in control of the horizontal and the vertical of the whole thing. We used our query parser to build a parse tree and that was used to build the list of must/may have terms we indexed the query under. We warned the customer away from using certain types of wildcards in the stored queries because it could cause an explosion in the number of queries selected.
Update for comment:
Short answer: I don't know for sure.
Longer answer: We were dealing with a custom built text search engine and part of it's query syntax allowed slicing the doc collection in certain ways very efficiently, with special emphasis on date_added. We played a lot of games because we were ingesting 4-10,000,000 new docs a day and running them against up to 1,000,000+ stored queries on a DEC Alphas with 64MB of main memory. (This was in the late 80's/early 90's.)
I'm guessing that filtering on something equivalent to date_added could be done used in combination the date of the last time you ran your queries, or maybe the highest id at last query run time. If you need to re-run the queries against a modified record you could use its id as part of the query.
For me to get any more specific, you're going to have to get a lot more specific about exactly what problem you are trying to solve and the scale of the solution you are trying accomplishing.
If you stored the type(s) of object(s) involved in each stored search as a generic relation, you could add a post-save signal to all involved objects. When the signal fires, it looks up only the searches that involve its object type and runs those. That probably will still run into scaling issues if you have a ton of writes to the db and a lot of saved searches, but it would be a straightforward Django approach.