# Primary key in DataFrame
**URL:** https://discourse.julialang.org/t/primary-key-in-dataframe/60879
**Category:** New to Julia
**Created:** [May 10, 2021, 11:43am UTC](https://discourse.julialang.org/t/primary-key-in-dataframe/60879 "2021-05-10T11:43:22Z")
**Posts on this page:** 3
**Page:** 1
### Author:  [@Mastomaki](https://discourse.julialang.org/u/Mastomaki)
#### Post date: [May 10, 2021, 11:43am UTC](https://discourse.julialang.org/t/primary-key-in-dataframe/60879/1 "2021-05-10T11:43:22Z")
I use DataFrame to hold data where a set of columns is considered the primary key. So only one record should be present for each primary key. But I suppose in DataFrame such condition cannot be enforced. So for example the following code where person and town would represent the primary key and time the queried value :
```julia
using DataFrames, DataFramesMeta
using Dates
df2 = DataFrame(person = ["John", "Nick", "Mary", "Mary"],
time = [DateTime("20150101","yyyymmdd"),
DateTime("20150101","yyyymmdd"),
DateTime("20150201","yyyymmdd"),
DateTime("20150601","yyyymmdd") ],
town = ["Brisbane", "Perth","Wollongong", "Wollongong"]);
e = df2[(df2.person .== "Mary") .& (df2.town .== "Wollongong"), :time]
show(e)
```
Suppose that person visiting the same town twice is impossible. I ask when Mary visited Wollongong and expect a **single date** but get an array if many records were originally put into df2.
Is it better to use a NamedArray or Dict instead of Dataframe to enforce records with unique primary keys? In Pandas there is `MultiIndex`, which I suppose can achieve something like this.
---
### Author:  [@nilshg](https://discourse.julialang.org/u/nilshg)
#### Post date: [May 10, 2021, 12:25pm UTC](https://discourse.julialang.org/t/primary-key-in-dataframe/60879/2 "2021-05-10T12:25:30Z")
Just `combine(groupby(df2, [:person, :town]), :time => first => :time)`, or whatever alternative function to decide which `time` entry you want to see.
You’re right that enforcing this doesn’t work in DataFrames, and you should probably use something else if this is of crucial importance. Deciding what exactly that might be will probably require more context - in your example you say that visiting the same town twice is impossible, but clearly in the data you posted it is possible, so one would need to consider why this happens to figure out what to do about it.
---
### Author:  [@rafael.guerra](https://discourse.julialang.org/u/rafael.guerra)
#### Post date: [May 10, 2021, 10:55pm UTC](https://discourse.julialang.org/t/primary-key-in-dataframe/60879/3 "2021-05-10T22:55:42Z")
Would it be useful to define a function that only adds a new row to the dataframe if this has unique (new) primary keys (person and town), otherwise the existing row would be overwritten?