I am trying to build excerpts for each document returned as a search results on my website. I am using the Sphinx search engine and the Apache web server on Linux CentOS. The function within the Sphinx API that I'd like to use is called BuildExcerpts. This function requires you to pass an array of strings where each string contains the documents contents.
I'm wondering what the best practice is for retrieving the document contents in real time as I serve the results on the web. Currently, these documents are in text files on my system, spread across multiple drives. There are roughly 100MM of them and they take up a few terabytes of space.
It's easy for me to call something like file_get_contents(), but that feels like the wrong way to do this. My databases are already gigantic ( 100GB+ ) and I don't particularly want to throw the document contents in there along with the document attributes that already exist. Perhaps this is the best way to do this, however.
Suggestions?
Well the source needs to be fetched from somewhere. If you dont want to duplicate it in your database, then you will need to fetch it from the filesystem. (using file_get_contets or similar)
Although the BuildExerpts function does give you one extra option "load_files"
... then sphinx will read the data from the filename for you.
What problem are you experiencing with reading it from files? Is it too slow? If so maybe use some caching in front - using memcache maybe.
Related
I am currently working on an Elixir/Phoenix project and I was wondering what is a good way to create a quick in-memory search index.
The index would be created on request and destroyed when the request is over and currently the data comes from a database via Ecto. Also, I would like to query it by different indexes so not just by :id but other indexes Example :user_id so a flat key value store may not be enough.
Are there any tools that would be helpful? I looked a bit into mnesia but when using it with ecto3_mnesia, a local file/folder was created and I would prefer if everything was in memory.
Thanks
I have no idea about ecto3_mnesia, but I am pretty sure raw :mnesia without any redundant wrapper is a good fit here (or, even, :ets if you don’t need a clustered solution.)
:mnesia.table_create/2 accepts many options, two you might be interested in are disc_copies and raw_copies. Simply initialize the former with empty node list and the latter with your complete node list, and you are all set: no disk copies are created, everything is in memory.
I am trying to store a plist and several binary files (let's say images) as part of an UIManagedDocument. The name of the binary files are an attribute in Core Data and I don't need to enumerate them, just access the right one when showing the related entity.
The file structure that I want to have is:
- <File yyyyMMdd-HHmmss>.extdoc
- StoreContent
- persistentStore
- AdditionalContent
- ListStatus.plist (used to store per document defaults)
- Images
- uuid1.png
- uuid2.png
- ...
- uuidn.png
So far, I have successfully followed the instructions in How do I save additional content into my UIManagedDocument file packages?, but when I try to add the binary files there are some things that I don't know how to do.
Should I treat the URL /the/path/File yyyyMMdd-HHmmss.extdoc/AdditionalContent (the default one provided with readAdditionalContentFromURL:error:) as a NSFileWrapper? Are there any advantages/disadvantages vs just using the URLs? I find it more complicated to use the file wrapper, since the plist has to be read using the file wrapper accessors and NSCoder (I guess), and the files, I have to store the file wrapper for the Images directory and then obtain the corresponding node with objectForKey (I assume). But Apple's Document-Based Apps Programming Guide for iOS regarding custom formats instead of NSData or NSFileWrapper, states "Keep in mind that your code will have to duplicate what UIDocument does for you, and so you must deal with greater complexity and a greater possibility of error." Am I misunderstanding this?
Per document defaults are declared as properties: the setter modifies the NSDictionary that maps the plist and marks the document as updated, and the getter accesses the dictionary with the proper key. How do I expose the ability to read/write the binary files? Should I add a method to my subclass of UIManagedDocument? - (void)writeImage:(NSString*)uuid; and -(UIImage *)readImage:(NSString *)uuid; And should I keep this data in memory until the document is saved? How?
Assuming that NSFileWrapper is the way to go, if I plan to use this document with iCloud should I use file coordinators with the file wrapper? If so, how?
Any source code for each question will be greatly appreciated. Thank you.
P.S.: I know that I could save some binary data inside of Core Data, but I don't feel comfortable with that solution. Among other reasons, I rather store the PNG data for image files that a serialized version of UIImage that won't be compatible with NSImage if I want to create a desktop app.
I'd like to say that, in general I rather like UIManagedDocument. It has a few advantages over raw Core Data. For example, it sets up the entire core data stack for you automatically. It also sets up nested managed object contexts for you, so you get free background saving. None of that is particularly earth-shattering, but it's a lot of functionality from a tiny amount of code.
I haven't played around with saving additional information...but here are my thoughts.
First, you shouldn't need to treat the new URL as a file wrapper. You should just be able to do regular file operations on the provided URL. Just make sure you have everything implemented properly in additionalContentForURL:error:, writeAdditionalContent:toURL:originalContentsURL:error: and readAdditionalContentFromURL:error:. The read and write operations need to be symmetric. And you should probably snapshot your data in additionalContentsForURL:error: so that everything will be saved in a known, good state (since the save operations are asynchronous).
As an alternative, have you considered using the Store in External Record File flag in your data model instead of saving it manually? This should force Core Data to (depending on the size of the binary data) automatically store them externally. I looked at the release notes, and I didn't see anything saying you couldn't use this feature with iCloud. That might be the easiest fix.
Attacking a side point for the moment (as I have not had ANY good experience with UIManagedDocument).
You can save the binary inside of Core Data for a iOS 5.0+ application using the external file reference. Then you can save the PNG of the image to Core Data directly and not need to worry about a UIManagedDocument or about bloating the sqlite file.
There is nothing stopping you from storing the PNG instead of a UIImage.
One other thought. You may need to use an NSFileCoordinator for the read and write operations. Technically, any read or write operations in the iCloud container need to use a file coordinator (to coordinate with the iCloud sync service--this prevents accidentally corrupting a file by reading it while another process is writing to it).
I know that UIDocument wraps most of its input and output methods automatically. I'd guess that these methods are similarly wrapped (since they give you a URL to use)--However, the docs aren't very clear.
Greetings,
in short, I have to find out whether I can implement a way to index zipped .rtf files via IFilter under Sql Server 2008 Express for fulltext search.
Long version:
this question is mostly a theoretical one - I'm neither experienced nor knowledgeable enough to find out whether such a thing is possible on my own.
The problem is as follows. There's a limited-size Sql Server Express 2008 R2 database thats going to store large .rtf files, probably 2-10k of them, and index them for fulltext search. Now, they probably wont fit into the 10gb limitation, thus I'm wondering if they could be archived (zipped, for instance) and stored that way. Fulltext search should be doable on them, in their zipped state.
My thought was to try to chain ifilters in some way to achieve this (I've no idea if thats doable), or there could be a different solution that I'm not seeing atm; I'd appreciate any input, as I'm kinda at a loss.
You may have a much easier time using something like Lucene. You could extract the text for the files and index it.
We own SiteA and SiteB and they share the same server and database where we have full control.
SiteC , siteD and siteE are some of the sites we own as well but reside on a different web hosts.
The goal is to create a unified search functionality for all of the sites mentioned above. That is if somebody search for a term in SiteA, the search result will automatically come up with results from SiteB,SiteC,SiteD and Site E too. The search results should be shown under the website they were found in.
All these websites content are stored in their own databases.
If I use SphinxSearch to index the above sites,I would then require those sites that we dont have complete control with to setup a web service where i can download a database dump or csv file for indexing.
Im not quite sure about how a sphider will come into play here so need your opinion.
Sphinx or a spider?
THanks!
If you can ask the owner of other websites to give you content for free, then there is no need for a spider. Just use sphinxsearch to index the content.
If you can't get content directly from them, a spider is the only choice for you. There is little to think about this issue.
Sphinx is a full-text search engine solution, while a spider is for fetching contents from internet. They are not replacements to each other. Even if you use a spider, you still have to use some full-text search engine software for example sphinx or lucene/solr.
So you have to make a decision first: Do I want to use sphinx for searching? If the answer is yes, then there is only one thing left: how can I index the contents for searching?
sphinx supports using database or XML as data source. Database as data source is more popular because preparing and updating XML documents in a specific format is very tedious(compared to maintaining a database table). So I guess finally you have to store all of the data into database. As you described, all of the data are all ready in databases, but some of the databases are out of your control. For you own database, there is no problem. For the databases that out of your control, I suggest that you use distributed sphinx searching: http://sphinxsearch.com/docs/2.0.6/distributed.html
The key idea is to horizontally partition (HP) searched data accross
search nodes and then process it in parallel.
Partitioning is done manually. You should
setup several instances of Sphinx programs (indexer and searchd) on
different servers;
make the instances index (and search) different parts of data;
configure a special distributed index on some of the searchd
instances;
and query this index.
This index only contains references to other local and remote indexes
- so it could not be directly reindexed, and you should reindex those indexes which it references instead.
I've got a system in which users select a couple of options and receive an image based on those options. I'm trying to combine multiple generated images(corresponding to those options) into the requested picture. I'm trying to optimize this so that if, an image exists for a certain option (i.e. the file exists), then there's no need to compute it and we move on to the next step.
Should I store these images in different folders, where each folder is an option name? Should I store them in the same folder, adding a prefix corresponding to the option to each image? Should I store the filenames in a database and check there? Which way is faster to check a file for existence?
I'm using PHP on Linux, but I'm also interested if the answer varies if I change the programming language or the OS.
If you're going to be producing a lot of these images, it doesn't seem very scalable to keep them all in one flat directory. I would go with a hierarchy, which will make it a lot easier to manage.
It's always going to be quicker to check in a database than to check if a file exists though, so if speed is the primary concern, use a hierarchical folder structure and keep all the filenames in a database.