Consider spreading attachment folders to subfolders to avoid 10000+ folders under a single root

[vlsi]

It looks like the current strategy is to have the structure like $JIRA_ID/..., however, you might have many issues, so it would make sense to spread the folders like $HASH/$JIRA_ID/..., so subfolders never get long. HASH could be the last two digits of the JIRA issue.

It would make manual navigation easier.

See https://github.com/vlsi/tmp-jmeter-attachments

[dweiss]

Do we care though? It's not like this is going to be checked out by anybody - these are just files served if people click on a link somewhere. And for unix systems the number of files hardly makes a big difference, I think?

[vlsi]

The change is trivial, and it would ease cases like "open repo in GitHub UI"

[vlsi]

Here's a recent "big folder causes sloness" issue in Ant: https://bz.apache.org/bugzilla/show_bug.cgi?id=66048

[dweiss]

The change is trivial but it leads to less intuitive final URLs. I read the ant issue and like I said - I don't think people will ever clone the migrated attachments repository - why would anybody? I think it's better to have more intuitive attachment URLs than yield to potential needs that are somewhat esoteric.

[vlsi]

Note that when https://github.com/apache/lucene-jira-archive/issues/127 is implemented, the URLs become far from human-readable.

markup:

<img src='https://vlsi.github.io/tmp-jmeter-attachments/48/42248/31902-undo.png'>

rendering:

Actual URL in the GitHub UI (just inspect the image above): https://camo.githubusercontent.com/6879b0ddd141b7fe2856a5d1c0a77d4495c3f0e48ae2e4f18ad3d21f20c40668/68747470733a2f2f766c73692e6769746875622e696f2f746d702d6a6d657465722d6174746163686d656e74732f34382f34323234382f33313930322d756e646f2e706e67

[dweiss]

Ok, fair enough. But for non-inlined links we'd still show the un-obfuscated URL, right? I honestly don't think the number of files in a folder matters much here but feel free to do as you wish.

[mocobeta]

We have other tasks and realistically speaking, there is only one developer who can work on this issue (me). I'd like to save implementation effort if the current directory structure does not cause any practical issues - could you please tell us what is the exact problem here? I don't go against the suggestion, I'm just saying the priority is low to me unless it's a really critical problem or other people want to pick this.

[vlsi]

GitHub limits the listing to 1000 entries only: https://github.com/DefinitelyTyped/DefinitelyTyped/tree/master/types