Query returns value that don't exist in PySpark Dataframe - dataframe

Question

Is there a way to create a subset dataframe from a dataframe and be sure that its values will be used afterward?
I have a huge PySpark Dataframe like this (simplified example):
id
timestamp
value
1
1658919602
5
1
1658919604
9
2
1658919632
2
Now I want to take a sample from it to test something, before running on the entire Dataframe. I get a sample by:
# Big dataframe
df = ...
# Create sample
df_sample = df.limit(10)
df_sample.show() shows some values.
Then I run this command, and sometimes it returns values that are present in df_sample and sometimes it returns values that are not present in df_sample but in df.
df_temp = df_sample.sort(F.desc('timestamp')).groupBy('id').agg(F.collect_list('value').alias('newcol'))
As if it's not using df_sample but picking in a non deterministic way 10 rows from df.
Interestingly, if I run df_sample.show() afterwards, it shows the same values as when it was first called.
Why is this happening?
Here's full code:
# Big dataframe
df = ...
# Create sample
df_sample = df.limit(10)
# shows some values
df_sample.show()
# run query
df_temp = df_sample.sort(F.desc('timestamp')).groupBy('id').agg(F.collect_list('value').alias('newcol')
# df_temp sometimes shows values that are present in df_sample, but sometimes shows values that aren't present in df_sample but in df
df_temp.show()
# Shows the exact same values as when it was first called
df_sample.show()
Edit1: I understand that Spark is lazy, but is there any way to force it to not be lazy in this scenario?

Answer 1

We can use sample function provided by spark to achieve this.Every time you run a sample() function it returns a different set of sampling records, To regenerate the same sample every time as you need to compare the results from your previous run. To get consistent same random sampling uses the same slice value for every run.
df=spark.range(100)
# Execute first time
print(df.sample(0.1,123).collect())
# Execute Second time with same seed-123
print(df.sample(0.1,123).collect())
# Execute with different seed-456
print(df.sample(0.1,456).collect())
Refer spark docs
Stratum sampling in spark

Answer 2

What worked was using df_sample = df.limit(10).cache() or df_sample = df.limit(10).persist(). Samkart's comment pointed me in this direction.