Support secondary Indexes
Need to be able to sort on something other than the rowkey
Restricting us to only the row key is very limiting. We currently have to maintain our own secondary indexes for other columns and this is very cumbersome, it should be a feature of the platform.
Azure Search service provides the ability to index and search data stored in Azure Tables. This functionality is currently available in public preview. You can learn about Azure Search here: https://azure.microsoft.com/services/search/, and about Azure Search Table indexer here: https://azure.microsoft.com/documentation/articles/search-howto-indexing-azure-tables/.
We will continue to evaluate building secondary indexing directly into Table Storage.
Steffen Gammelgård commented
Hope the secondary index is still coming.
Azure Table Storage is an immensely powerful platform, and is already an exceptionally cost effective NoSQL database for certain kinds of applications.
Being able to add a secondary index would drastically reduce the complexity and the amount of redundant data in certain scenarios and would make an already awesome platform relevant for even more projects.
I too have used index tables to get secondary indicies and sometimes that would still be necessary to allow for efficient partitioning, but having a second index still seems like it would be very useful!
While I agree that built-in secondary indexes would be nice, in most cases it's better to consider indexes on a case-by-case basis and implement the index differently depending on your queries and their frequency. Information on creating your own secondary indexes can be found here https://msdn.microsoft.com/en-us/library/dn589791.aspx
+1 on Inattention and Neglect.
And with the evolution of discussion on this, it has become clear that to base critical software on top of Azure Table Storage with any needs beyond simple distributed partially sorted Hashtable is pointless; indeed, I can no longer trust UV as a channel where roadmap about these features is actually accurately communicated.
No, AIS is not a solution to this. I do not want an index that may, or may not be in an updated state. Heck, I am even willing to build my own index! Just let me grab multiple non-contiguous rows (by RowKey) in one request!
But more importantly than even that, decide if UserVoice is actually a channel that you want to use to engage with your customers. If it is, announcing at your conference that a feature is coming, then ignoring it for 2 years, then another 3 years before 2 years after that saying it actually is not coming, more than likely, is incredibly insulting.
inattention and neglect is the perfect summary of ATS. Microsoft, up the price or do whatever, just do something.
SSD-backed storage + secondary indexes is a no-brainer.
Appreciate the workaround. But the consensus that ATS has become - through inattention and neglect - an also-ran is hard to shake. Not sure how I could recommend that anyone build a solution on it at this point. At most, I would recommend ATS as a read-only data store, where you put data that you hope to never see again but aren't quite willing to say goodbye to.
Aaron, DocumentDB is not a replacement, it is a totally different beast and not a key-value store such as Table Storage. Also DocumentDB is not portable (yet at least) and gives you a lock-in to the public Azure cloud. Table Storage is portable through Azure Stack.
I dream and hope MS is picking up Table Storage again and gives us Premium (SSD-backed) and with secondary indexes. There's really no reason on this good earth they shouldn't do this!
Aaron Lawrence commented
I would agree with Mike Olsen that it is a BAD IDEA to build a new application on Azure Table Storage, as Microsoft appear to be deprecating it (although it's not completely clear what they regard as a replacement, DocumentDB seems most likely)
Regarding table storage, I've also noticed that in the new portal when you look at diagnostics config in WebApps, there's only File Storage and Blob Storage as options, while in the classic portal you also have Table Storage as an option.
Even more, I've noticed that the internal logging in WebJobs via WebJobs Dashboard seems to have moved from using Table Storage to using Blob Storage - but that may be for any number of reasons, of course.
Anyways, I wouldn't recommend table storage as a long-term option for anything new for these reasons. And of course it's hard to use without any secondary indexing.
Mike Olson commented
The writing's on the wall folks. It's been what, 3 years since this was supposedly added to their roadmap? When was the last time Microsoft made *any* real improvements to Table Storage, besides putting a ton of breaking changes in the SDK that required hundreds of hours of dev to support?
Azure Table Storage is old news and is being ushered unceremoniously out the door. It's time for us to give in and move to DocumentDB, Azure SQL, or one of the dozen new storage services that Microsoft is actually putting effort into. Or just do what I've been doing and just start storing your data in flat files on Blob Storage, since for a lot of cases that's easier to code, more performant and more flexible.
At least Microsoft isn't slowly moving us away from Cloud Services... oh wait.
Any comments from Microsoft admins ?
Aaron Lawrence commented
It's difficult to use ATS as it stands for anything real. We built our own indices in SQL, which of course introduces exciting new consistency issues. One index is just too little - basically that gives you the ability to move data in and out by an identifier, but not do anything else with it.
We absolutely need this! Due to the high cost and throughput limitations of DocumentDB it is not a viable alternative to having secondary indexes in table storage.
Let me give you an example of why this is critical. I have about 30 GB of event logs. I want to store these in ATS. The problem is that I need to be able to query them along multiple different axes - the company they belong to, the individual user they belong to, the date/time they came in, the event name, and so forth. And eventually, probably lots more. My current solution is to store the data "n" times, each one with a different partition-key/row-key schema, to enable querying along that particular dimension. So far so good - it's not a problem to write the data "n" times, given how well ATS performs, and how cheap it is.
But the problem is maintenance. Right now, with about 30 GB of data, if I come up with a new dimension that I need to support, it takes me at least a day to write the scripts to export and then re-import the data to the new format (because it all has to be parallelized, and needs to track state for each portion I'm running in parallel, or it would take weeks); and even if I don't ***** up on the (very complex) import/export scripts somehow, it then takes at least 1-2 days to actually get all the data over to the new table.
And that's simply not a scalable model. What happens when I don't have 30 GB of data I need to pivot, but 300 GB? Or 30 TB? The pivot scripts will take weeks to run, and maintaining any shred of consistency through the process gets very, very complicated.
I get that this is a complex problem to solve. But it doesn't make it any less complicated by telling every ATS user to come up with their own (almost certainly unoptimal and buggy) solution.
With the DocumentDB and Search services now in play, does Azure still plan to add secondary index support to Table Storage?
Yet another 6 months passed since we last heard from you MS
You can manage this by hand, I do it today and built some reusable code to maintain this. However, changing or adding indexes in the future is a pain, and with the lack of cross table transactions, I have had instances where an index did not get created. Which means if you want/need true integrity, you need to have other process/patterns in place.
So while, yeah you can do this today, other systems, even Amazon's does this now for you. I would love to not have to deal with this.
I wonder the same thing...at first glance this feature seems like an essential key ingredient, but on further examination the 'feature' can be implemented as part of the table design.
Am I right in saying that the only real benefit this feature would bring would be automatic guarantees of integrity across multiple entries? So the programmer would not need to rely on azure queues for example?
I would agree that this should be a high priority feature...but largely for non-technical reasons. Am I correct in understanding that the real problem here is a development community entrenched in the relational mindset, and who need features like this to drive adoption and improve the reputation of the platform?
John Leidegren commented
Does this even make sense? I thought the point was that you'd build out any secondary indexes much in the sense that you have to think of it as a asynchronous process that eventually indexes all your data. What else could you possibly be doing at this scale?
What Anders Madsen wrote in his comment makes a lot of sense to me.
Anders Madsen commented
Price kept in mind, I think that ATS functions and performs very well, and there's no reason why you can't index the data yourself.
I've used a radical approach, where I create Btree indexes, saved in blobs. I then use cheap small workerroles that executes SQL-like queries, and it performs very well.
Only thing to keep in mind, is that this approach creates stale indexes, so it primarily targets applications that access the table storage in an asynchronous way, e.g. reading and writing at different times of the day.
Shaun Tonstad commented
@Jeff Windows Azure Storage / Tables is something of a joke among savy cloud developers. Four years have past since this original request was made. There is no sense of urgency from MS to make this service competitive with other solutions (i.e. AWS SimpleDb). It's clear that MS lacks the expertise to improve it or the political desire given dominance of SQL Azure.