-
Notifications
You must be signed in to change notification settings - Fork 159
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
[DOCS] Partitioning user guide and small doc fixes #2717
Conversation
CodSpeed Performance ReportMerging #2717 will not alter performanceComparing Summary
|
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Looks great! just had two small comments on the wording
|
||
1. **Have Enough Partitions**: our general recommendation for high throughput and maximal resource utilization is to have *at least* ``2 x TOTAL_NUM_CPUS`` partitions, which allows Daft to fully saturate your CPUs. | ||
2. **More Partitions**: if you are observing memory issues (excessive spilling or out-of-memory (OOM) issues) then you may choose to increase the number of partitions. This increases the amount of overhead in your system, but improves overall memory stability (since each partition will be smaller). | ||
3. **Fewer Partitions**: if you are observing a large amount of overhead (especially during shuffle operations such as joins and sorts), then you may choose to decrease the number of partitions. This decreases the amount of overhead in the system, at the cost of using more memory (since each partition will be larger). |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Maybe some description of how to measure overhead (vs maybe just an operation that is expensive)?
--------------------------- | ||
|
||
Daft will automatically use certain heuristics to determine the number of partitions for you when you create a DataFrame. When reading data from files (e.g. Parquet, CSV or JSON), | ||
each file is by default one partition on its own, but Daft will also perform splitting of partitions (for files that are egregiously large) and coalescing of partitions (for small files) |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I think "by default" is a little misleading here as Daft by default does scan task splitting and merging.
Closes: #840