I've over 5 million key-value pairs stored in REDIS server. Incoming key, which we need to search in REDIS server will be like "key_num_id" format.
Keys and values are stored in the REDIS server like "key_pfx_id" format. All keys will be unique key (No two keys would be same). Below are few examples:
key_1234_11
key_123_12
key_123_11
key_12_11
..
..
where 1234, 123, 12 are the prefix of num in incoming key key_num_id.
Now, for example if we get key_1234567890_11 as the incoming key then REDIS should give value corresponding to "key_1234_11" which is the best match for the "num" we got in incoming key "1234567890" in our example.
One way is to do this is query the REDIS server multiple times till we got the value; e.g.
GET key_1234567890_11
GET key_123456789_11
GET key_12345678_11
GET key_1234567_11
.
.
But I think this is the costly solution as I'm getting around 2000 incoming keys in a second. So want to have optimized solution. Can anyone help in this as I'm newbie in REDIS
NOTE: I'm doing all above in C code
One way you could do this in a single call to Redis is put the iteration logic in a Lua script.
local input = KEYS[1]
local key, output
while string.len(input) > 1 do
key = "test_" .. input .. "_11"
output = redis.call("GET", key)
if output then return output end
input = string.sub(input, 1, string.len(input) - 1)
end
return nil
When called (redis.eval(script_body, ["12345678"])) it should correctly return the value of key_1234_11. This is a trivial example, obviously, and will need to be further tailored for your application.
This saves you the overhead of making many calls to redis, but, I guess compared to C then Lua itself might be a bit of an overhead. I can't be certain a Lua script is faster in this scenario, but it may be.
Related
I need to cluster some data in Snowflake using DBSCAN. I created a UDF but the results won't match with a local run, so I ran an UDF that just creates a list with the row number that is being processed and it results in a list that has repeated values and its max value is much smaller than the number of rows in my table. (The expected result was unique values up to the number of rows)
Can this be a parallelization issue?
If so, is there a way to cluster data using DBSCAN in Snowflake?
Thanks!
EDIT -> code example:
#funcs.pandas_udf(name='DBSCAN_TEST', is_permanent=True,
stage_location='#UDF', replace=True,
packages=['scikit-learn==1.0.2', 'pandas', 'numpy'])
def DBSCAN_TEST(data_x: types.PandasDataFrame[float, float]) -> types.PandasSeries[float]:
data_x.columns = [features]
DBSCAN_cluster = DBSCAN(eps=2.5, min_samples=4)
DBSCAN_cluster.fit(data_x)
return resul
As input data I used this
to test DBSCAN.
If I run that locally (using the exact same code inside the UDF) I end up with 24 clusters and that is the expected result. But if I use the UDF it scales up to 71.
I've tried changing the input types to string, as a coworker suggested but it didn't work.
The only clustering option in Snowflake is a clustering key, and general rule of thumb is that this isn't something typically needed until you eclipse a TB of data in a table and/or the auto clustering proves to be deficient in performance.
https://docs.snowflake.com/en/user-guide/tables-clustering-keys.html
After working with neo4j and now coming to the point of considering to make my own entity manager (object manager) to work with the fetched data in the application, i wonder about neo4j's output format.
When i run a query it's always returned as tabular data. Why is this??
Sure tables keep a big place in data and processing, but it seems so strange that a graph database can only output in this format.
Now when i want to create an object graph in my application i would have to hydrate all the objects and this is not really good for performance and doesn't leverage true graph performace.
Consider MATCH (A)-->(B) RETURN A, B when there is one A and three B's, it would return:
A B
1 1
1 2
1 3
That's the same A passed down 3 times over the database connection, while i only need it once and i know this before the data is fetched.
Something like this seems great http://nigelsmall.com/geoff
a load2neo is nice, a load-from-neo would also be nice! either in the geoff format or any other formats out there https://gephi.org/users/supported-graph-formats/
Each language could then implement it's own functions to create the objects directly.
To clarify:
Relations between nodes are lost in tabular data
Redundant (non-optimal) format for graphs
Edges (relations) and vertices (nodes) are usually not in the same table. (makes queries more complex?)
Another consideration (which might deserve it's own post), what's a good way to model relations in an object graph? As objects? or as data/method inside the node objects?
#Kikohs
Q: What do you mean by "Each language could then implement it's own functions to create the objects directly."?
A: With an (partial) graph provided by the database (as result of a query) a language as PHP could provide a factory method (in C preferably) to construct the object graph (this is usually an expensive operation). But only if the object graph is well defined in a standard format (because this function should be simple and universal).
Q: Do you want to export the full graph or just the result of a query?
A: The result of a query. However a query like MATCH (n) OPTIONAL MATCH (n)-[r]-() RETURN n, r should return the full graph.
Q: you want to dump to the disk the subgraph created from the result of a query ?
A: No, existing interfaces like REST are prefered to get the query result.
Q: do you want to create the subgraph which comes from a query in memory and then request it in another language ?
A: no i want the result of the query in another format then tabular (examples mentioned)
Q: You make a query which only returns the name of a node, in this case, would you like to get the full node associated or just the name ? Same for the edges.
A: Nodes don't have names. They have properties, labels and relations. I would like enough information to retrieve A) The node ID, it's labels, it's properties and B) the relation to other nodes which are in the same result.
Note that the first part of the question is not a concrete "how-to" question, rather "why is this not possible?" (or if it is, i like to be proven wrong on this one). The second is a real "how-to" question, namely "how to model relations". The two questions have in common that they both try to find the answer to "how to get graph data efficiently in PHP."
#Michael Hunger
You have a point when you say that not all result data can be expressed as an object graph. It reasonable to say that an alternative output format to a table would only be complementary to the table format and not replacing it.
I understand from your answer that the natural (rawish) output format from the database is the result format with duplicates in it ("streams the data out as it comes"). I that case i understand that it's now left to an alternative program (of the dev stack) to do the mapping. So my conclusion on neo4j implementing something like this:
Pro's - not having to do this in every implementation language (of the application)
Con's - 1) no application specific mapping is possible, 2) no performance gain if implementation language is fast
"Even if you use geoff, graphml or the gephi format you have to keep all the data in memory to deduplicate the results."
I don't understand this point entirely, are you saying that these formats are no able to hold deduplicated results (in certain cases)?? So infact that there is no possible textual format with which a graph can be described without duplication??
"There is also the questions on what you want to include in your output?"
I was under the assumption that the cypher language was powerful enough to specify this in the query. And so the output format would have whatever the database can provide as result.
"You could just return the paths that you get, which are unique paths through the graph in themselves".
Useful suggestion, i'll play around with this idea :)
"The dump command of the neo4j-shell uses the approach of pulling the cypher results into an in-memory structure, enriching it".
Does the enriching process fetch additional data from the database or is the data already contained in the initial result?
There is more to it.
First of all as you said tabular results from queries are really commonplace and needed to integrate with other systems and databases.
Secondly oftentimes you don't actually return raw graph data from your queries, but aggregated, projected, sliced, extracted information out of your graph. So the relationships to the original graph data are already lost in most of the results of queries I see being used.
The only time that people need / use the raw graph data is when to export subgraph-data from the database as a query result.
The problem of doing that as a de-duplicated graph is that the db has to fetch all the result data data in memory first to deduplicate, extract the needed relationships etc.
Normally it just streams the data out as it comes and uses little memory with that.
Even if you use geoff, graphml or the gephi format you have to keep all the data in memory to deduplicate the results (which are returned as paths with potential duplicate nodes and relationships).
There is also the questions on what you want to include in your output? Just the nodes and rels returned? Or additionally all the other rels between the nodes that you return? Or all the rels of the returned nodes (but then you also have to include the end-nodes of those relationships).
You could just return the paths that you get, which are unique paths through the graph in themselves:
MATCH p = (n)-[r]-(m)
WHERE ...
RETURN p
Another way to address this problem in Neo4j is to use sensible aggregations.
E.g. what you can do is to use collect to aggregate data per node (i.e. kind of subgraphs)
MATCH (n)-[r]-(m)
WHERE ...
RETURN n, collect([r,type(r),m])
or use the new literal map syntax (Neo4j 2.0)
MATCH (n)-[r]-(m)
WHERE ...
RETURN {node: n, neighbours: collect({ rel: r, type: type(r), node: m})}
The dump command of the neo4j-shell uses the approach of pulling the cypher results into an in-memory structure, enriching it and then outputting it as cypher create statement(s).
A similar approach can be used for other output formats too if you need it. But so far there hasn't been the need.
If you really need this functionality it makes sense to write a server-extension that uses cypher for query specification, but doesn't allow return statements. Instead you would always use RETURN *, aggregate the data into an in-memory structure (SubGraph in the org.neo4j.cypher packages). And then render it as a suitable format (e.g. JSON or one of those listed above).
These could be a starting points for that:
https://github.com/jexp/cypher-rs
https://github.com/jexp/cypher_websocket_endpoint
https://github.com/neo4j-contrib/rabbithole/blob/master/src/main/java/org/neo4j/community/console/SubGraph.java#L123
There are also other efforts, like GraphJSON from GraphAlchemist: https://github.com/GraphAlchemist/GraphJSON
And the d3 json format is also pretty useful. We use it in the neo4j console (console.neo4j.org) to return the graph visualization data that is then consumed by d3 directly.
I've been working with neo4j for a while now and I can tell you that if you are concerned about memory and performances you should drop cypher at all, and use indexes and the other graph-traversal methods instead (e.g. retrieve all the relationships of a certain type from or to a start node, and then iterate over the found nodes).
As the documentation says, Cypher is not intended for in-app usage, but more as a administration tool. Furthermore, in production-scale environments, it is VERY easy to crash the server by running the wrong query.
In second place, there is no mention in the docs of an API method to retrieve the output as a graph-like structure. You will have to process the output of the query and build it.
That said, in the example you give you say that there is only one A and that you know it before the data is fetched, so you don't need to do:
MATCH (A)-->(B) RETURN A, B
but just
MATCH (A)-->(B) RETURN B
(you don't need to receive A three times because you already know these are the nodes connected with A)
or better (if you need info about the relationships) something like
MATCH (A)-[r]->(B) RETURN r
Does anybody have experience with getting random results from index with +100,000,000 (100 million) records.
The goal is getting 30 results ordered by random, at least 100 times per second.
Actually my records are in MySQL but selecting ORDER BY RAND() from huge tables is the most easiest way to kill MySQL.
Sphinxsearch or whatever what do you recommend?
I dont have that big an index to try.
barry#server:~/modules/sphinx-2.0.1-beta/api# time php test.php -i gi_stemmed --sortby #random --select id
Query '' retrieved 20 of 3067775 matches in 0.081 sec.
Query stats:
Matches:
<SNIP>
real 0m0.100s
user 0m0.010s
sys 0m0.010s
This is on a reasonably powerful dedicated server - that is serving live queries (~20qps)
But to be honest if you dont need filtering (ie each query has a 'WHERE' clause), you can just setup a system that returns random results - can do this with mysql. Just using ORDER BY RAND() is evil (and sphinx while better at sorting than mysql is still doing basically the same thing).
How 'sparse' is your data? If most of your ids are used, can just do soemthing like
$ids = array();
$max = getOne("SELECT MAX(id) FROM table");
foreach(range(1,30) as $idx) {
$ids[] = rand(1,$max);
}
$query = "SELECT * FROM table WHERE id IN (".implode(',',$ids).")";
(may want to use shuffle() in php on the results afterwards as you likly to get the results out of mysql in id order)
Which will be much more efficient. If you do have holes, perhaps just lookup 33 rows. Sometimes will get more than need, (just discard), but you should still get 30 most of the times.
(Of course you could cache the '$max' somewhere, so it doesnt have to be looked up all the time.)
Otherwise you could setup a dedicated 'shuffled' list. Basically a FIFO buffer, have one thread, filling it with random results (perhaps using the above system, using 3000 ids at a time) and then the consumers just read random results directly out of this queue.
FIFO, is not particully easy to implement with mysql, so maybe use a different system - maybe redis, or even just memcache.
I have a set of entries in the datastore and I would like to search/retrieve them as user types query. If I have full string it's easy:
q := datastore.NewQuery("Products").Filter("Name =", name).Limit(20)
but I have no idea how to do it with partial string, please help.
q := datastore.NewQuery("Products").Filter("Name >", name).Limit(20)
There is no like operation on app engine but instead you can use '<' and '>'
example:
'moguz' > 'moguzalp'
EDIT: GAH! I just realized that your question is Go-specific. My code below is for Python. Apologies. I'm also familiar with the Go runtime, and I can work on translating to Python to Go later on. However, if the principles described are enough to get you moving in the right direction, let me know and I wont' bother.
Such an operation is not directly supported on the AppEngine datastore, so you'll have to roll your own functionality to meet this need. Here's a quick, off-the-top-of-my-head possible solution:
class StringIndex(db.Model):
matches = db.StringListProperty()
#classmathod
def GetMatchesFor(cls, query):
found_index = cls.get_by_key_name(query[:3])
if found_index is not None:
if query in found_index.matches:
# Since we only query on the first the characters,
# we have to roll through the result set to find all
# of the strings that matach query. We keep the
# list sorted, so this is not hard.
all_matches = []
looking_at = found_index.matches.index(query)
matches_len = len(foundIndex.matches)
while start_at < matches_len and found_index.matches[looking_at].startswith(query):
all_matches.append(found_index.matches[looking_at])
looking_at += 1
return all_matches
return None
#classmethod
def AddMatch(cls, match) {
# We index off of the first 3 characters only
index_key = match[:3]
index = cls.get_or_insert(index_key, list(match))
if match not in index.matches:
# The index entity was not newly created, so
# we will have to add the match and save the entity.
index.matches.append(match).sort()
index.put()
To use this model, you would need to call the AddMatch method every time that you add an entity that would potentially be searched on. In your example, you have a Product model and users will be searching on it's Name. In your Product class, you might have a method AddNewProduct that creates a new entity and puts it into the datastore. You would add to that method StringIndex.AddMatch(new_product_name).
Then, in your request handler that gets called from your AJAXy search box, you would use StringIndex.GetMatchesFor(name) to see all of the stored products that begin with the string in name, and you return those values as JSON or whatever.
What's happening inside the code is that the first three characters of the name are used for the key_name of an entity that contains a list of strings, all of the stored names that begin with those three characters. Using three (as opposed to some other number) is absolutely arbitrary. The correct number for your system is dependent on the amount of data that you are indexing. There is a limit to the number of strings that can be stored in a StringListProperty, but you also want to balance the number of StringIndex entities that are in your datastore. A little bit of math with give you a reasonable number of characters to work with.
If the number of keywords is limited you could consider adding an indexed list property of partial search strings.
Note that you are limited to 5000 indexes per entity, and 1MB for the total entity size.
But you could also wait for Cloud SQL and Full Text Search API to be avaiable for the Go runtime.
I've been using SQL Server to store historical time series data for a couple hundred thousand objects, observed about 100 times per day. I'm finding that queries (give me all values for object XYZ between time t1 and time t2) are too slow (for my needs, slow is more then a second). I'm indexing by timestamp and object ID.
I've entertained the thought of using somethings a key-value store like MongoDB instead, but I'm not sure if this is an "appropriate" use of this sort of thing, and I couldn't find any mentions of using such a database for time series data. ideally, I'd be able to do the following queries:
retrieve all the data for object XYZ between time t1 and time t2
do the above, but return one date point per day (first, last, closed to time t...)
retrieve all data for all objects for a particular timestamp
the data should be ordered, and ideally it should be fast to write new data as well as update existing data.
it seems like my desire to query by object ID as well as by timestamp might necessitate having two copies of the database indexed in different ways to get optimal performance...anyone have any experience building a system like this, with a key-value store, or HDF5, or something else? or is this totally doable in SQL Server and I'm just not doing it right?
It sounds like MongoDB would be a very good fit. Updates and inserts are super fast, so you might want to create a document for every event, such as:
{
object: XYZ,
ts : new Date()
}
Then you can index the ts field and queries will also be fast. (By the way, you can create multiple indexes on a single database.)
How to do your three queries:
retrieve all the data for object XYZ
between time t1 and time t2
db.data.find({object : XYZ, ts : {$gt : t1, $lt : t2}})
do the above, but return one date
point per day (first, last, closed to
time t...)
// first
db.data.find({object : XYZ, ts : {$gt : new Date(/* start of day */)}}).sort({ts : 1}).limit(1)
// last
db.data.find({object : XYZ, ts : {$lt : new Date(/* end of day */)}}).sort({ts : -1}).limit(1)
For closest to some time, you'd probably need a custom JavaScript function, but it's doable.
retrieve all data for all objects for
a particular timestamp
db.data.find({ts : timestamp})
Feel free to ask on the user list if you have any questions, someone else might be able to think of an easier way of getting closest-to-a-time events.
This is why databases specific to time series data exist - relational databases simply aren't fast enough for large time series.
I've used Fame quite a lot at investment banks. It's very fast but I imagine very expensive. However if your application requires the speed it might be worth looking it.
There is an open source timeseries database under active development (.NET only for now) that I wrote. It can store massive amounts (terrabytes) of uniform data in a "binary flat file" fashion. All usage is stream-oriented (forward or reverse). We actively use it for the stock ticks storage and analysis at our company.
I am not sure this will be exactly what you need, but it will allow you to get the first two points - get values from t1 to t2 for any series (one series per file) or just take one data point.
https://code.google.com/p/timeseriesdb/
// Create a new file for MyStruct data.
// Use BinCompressedFile<,> for compressed storage of deltas
using (var file = new BinSeriesFile<UtcDateTime, MyStruct>("data.bts"))
{
file.UniqueIndexes = true; // enforces index uniqueness
file.InitializeNewFile(); // create file and write header
file.AppendData(data); // append data (stream of ArraySegment<>)
}
// Read needed data.
using (var file = (IEnumerableFeed<UtcDateTime, MyStrut>) BinaryFile.Open("data.bts", false))
{
// Enumerate one item at a time maxitum 10 items starting at 2011-1-1
// (can also get one segment at a time with StreamSegments)
foreach (var val in file.Stream(new UtcDateTime(2011,1,1), maxItemCount = 10)
Console.WriteLine(val);
}
I recently tried something similar in F#. I started with the 1 minute bar format for the symbol in question in a Space delimited file which has roughly 80,000 1 minute bar readings. The code to load and parse from disk was under 1ms. The code to calculate a 100 minute SMA for every period in the file was 530ms. I can pull any slice I want from the SMA sequence once calculated in under 1ms. I am just learning F# so there are probably ways to optimize. Note this was after multiple test runs so it was already in the windows Cache but even when loaded from disk it never adds more than 15ms to the load.
date,time,open,high,low,close,volume
01/03/2011,08:00:00,94.38,94.38,93.66,93.66,3800
To reduce the recalculation time I save the entire calculated indicator sequence to disk in a single file with \n delimiter and it generally takes less than 0.5ms to load and parse when in the windows file cache. Simple iteration across the full time series data to return the set of records inside a date range in a sub 3ms operation with a full year of 1 minute bars. I also keep the daily bars in a separate file which loads even faster because of the lower data volumes.
I use the .net4 System.Runtime.Caching layer to cache the serialized representation of the pre-calculated series and with a couple gig's of RAM dedicated to cache I get nearly a 100% cache hit rate so my access to any pre-computed indicator set for any symbol generally runs under 1ms.
Pulling any slice of data I want from the indicator is typically less than 1ms so advanced queries simply do not make sense. Using this strategy I could easily load 10 years of 1 minute bar in less than 20ms.
// Parse a \n delimited file into RAM then
// then split each line on space to into a
// array of tokens. Return the entire array
// as string[][]
let readSpaceDelimFile fname =
System.IO.File.ReadAllLines(fname)
|> Array.map (fun line -> line.Split [|' '|])
// Based on a two dimensional array
// pull out a single column for bar
// close and convert every value
// for every row to a float
// and return the array of floats.
let GetArrClose(tarr : string[][]) =
[| for aLine in tarr do
//printfn "aLine=%A" aLine
let closep = float(aLine.[5])
yield closep
|]
I use HDF5 as my time series repository. It has a number of effective and fast compression styles which can be mixed and matched. It can be used with a number of different programming languages.
I use boost::date_time for the timestamp field.
In the financial realm, I then create specific data structures for each of bars, ticks, trades, quotes, ...
I created a number of custom iterators and used standard template library features to be able to efficiently search for specific values or ranges of time-based records.