I am trying to parse Android notification logs into a database. The columns are uneven because different notifications omit certain attributes. Essentially, it is a CSV file of key-value pairs. To put it another way, it is a really poorly formatted (& syntactically incorrect) JSON file. Maybe it would be easier to show you...


I have truncated them for the sake of brevity, but they do not line up toward the end of the rows. Some text values are quoted, some are not. All of the keys are quoted. I can use string functions to trim the extraneous characters once I get the columns sorted, but therein lies the rub: How do I sort the values into columns? I can do this in about 4 different programming languages, but I am just learning T-SQL.

How can I parse this to SQL Server (2016) in a single query? Is it even possible?

I could import it to a temporary table & then transfer it to the permanent one, but I would like to have one query to digest the info, since this will be repeated on a daily (maybe more) basis. The magic would be if I could declare a second-level delimiter, since all values are preceded by a colon. Declaring the colon as the delimiter breaks it the other way...arg(s)!

The data is all around ragged. Even the notifications from the same app may be drastically different. The way I would do it in PHP, C, JS, etc would be to loop through the keys until I hit the ones I needed, grab the appropriate values following them (or the whole pair), then trim the extraneous chars. I just don't know how to do it in T-SQL & the online docs leave something to be desired.

The expected result would be something like this:

Package Name | Post Time | System Time | Offset | NotificationText
automatic    | 1489795   | 786083      | -1400  | this is the note
gameservice  | 1489756   | 786090      | -1400  | this is the note
automatic    | 1489799   | 786045      | -1400  | this is the note

The big problem is that the info I need is scattered throughout the rows. I don't need much of it, but it is buried.

  • Yeah, don't do that with T-SQL. Process your text files and import the cleaned versions. Mar 22 '17 at 1:21
  • @sp_BlitzErik Not possible, or prohibitively complicated? The goal is to eliminate steps from the workflow. I can code things that are complicated, I just don't know T-SQL very intimately, & I can't seem to figure out the conditional logic & string matching functions. Thanks for the help all the same though! I totally appreciate it.
    – AVLien
    Mar 22 '17 at 1:30
  • 1
    I think what @sp_BlitzErik is trying to say is that SQL Server is a ridiculously expensive way to do row-by-row string parsing. Best practice is to do the parsing in a procedural language before you insert it into the database table(s) and leave the database server to do database stuff, like storage, sorting, aggregating. Mar 22 '17 at 13:18
  • 2
    Hahaha, I fell asleep before I could respond last night. But yeah, @DanielHutmacher is right. I've tried to do stuff like this early on in my T-SQL days as well, and it was so bad I resorted to doing it with PowerShell instead. PowerShell! That's how bad it was. Mar 22 '17 at 13:44

You can use the technique shown in Loading line-delimited JSON files in SQL Server 2016 on the SQL Server Database Engine Blog, written by Jovan Popovic .

For example, given a file import.txt:


and a format file import.fmt:

1 SQLCHAR 0 0 "\r\n" 1 json ""

The file can be bulk imported and parsed with:

    BULK 'C:\Temp\import.txt',
    FORMATFILE= 'C:\Temp\import.fmt'
) AS BulkLog
    packageName varchar(50),
    postTime    bigint,
    systemTime  bigint,
    offset      bigint,
    isOngoing   bit
) AS Parsed;


Parsed results


Not sure why you think this is syntactically incorrect JSON, but according to json.org, it looks fine. Well, except for 2 minor things:

  1. need commas between sets of { and }
  2. Need [ and ] around the entire thing

But outside of that, it is supposed to be that all keys are quoted, and that numbers and booleans values are not quoted.

SQL Server 2016 has built-in JSON parsing. After fixing those two minor issues with the sample data in the question, the below parses it just fine:


   SELECT [PackageName],
   WITH (
      [PackageName] VARCHAR(MAX)       '$.packageName',
      [IsOngoing]   BIT                '$.isOngoing'


PackageName                IsOngoing
com.automatic.mustang      0
com.enhance.gameservice    0
com.automatic.mustang      0
com.enhance.gameservice    0

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.