Yes, a single file compressed with a non-splitable compression (e.g.
gzip) would have to be read by a single executor. That takes forever.
You should consider to recompress the file with a splitable compression
first. You will not want to read that file more than once, so you should
uncompress it only once (in order to recompress).
Enrico
Am 22.06.22 um 20:17 schrieb Sid:
Hi Enrico,
Thanks for the insights.
Could you please help me to understand with one example of compressed
files where the file wouldn't be split in partitions and will put load
on a single partition and might lead to OOM error?
Thanks,
Sid
On Wed, Jun 22, 2022 at 6:40 PM Enrico Minack <i...@enrico.minack.dev>
wrote:
The RAM and disk memory consumtion depends on what you do with the
data after reading them.
Your particular action will read 20 lines from the first partition
and show them. So it will not use any RAM or disk, no matter how
large the CSV is.
If you do a count instead of show, it will iterate over the each
partition and return a count per partition, so no RAM here needed
as well.
If you do some real processing of the data, the requirement RAM
and disk again depends on involved shuffles and intermediate
results that need to be store in RAM or on disk.
Enrico
Am 22.06.22 um 14:54 schrieb Deepak Sharma:
It will spill to disk if everything can’t be loaded in memory .
On Wed, 22 Jun 2022 at 5:58 PM, Sid <flinkbyhe...@gmail.com> wrote:
I have a 150TB CSV file.
I have a total of 100 TB RAM and 100TB disk. So If I do
something like this
spark.read.option("header","true").csv(filepath).show(false)
Will it lead to an OOM error since it doesn't have enough
memory? or it will spill data onto the disk and process it?
Thanks,
Sid
--
Thanks
Deepak
www.bigdatabig.com <http://www.bigdatabig.com>
www.keosha.net <http://www.keosha.net>