SharedDirectoryListResponse

[ibeckermayer]

Adds listContents to get the information about each item in a directory and send it back in the fsoList of the SharedDirectoryListResponse.

I initially tried to create a version of listContents which gathered all of the promises in a list and then returned them wrapped in a Promise.all. However with this structure, sharing a directory with a large number of items (like the node_modules directory of this repo after yarn install) would cause the system to freeze up for a while.

I think its a similar problem to the one I encountered way back when working on the display pipeline, namely that piling up a lot of asynchronous calls clogs up the microtask queue and can make performance appear worse, because new regular tasks can't be executed while all the microtasks are processed.

Performance still isn't fantastic when sharing a large directory, but its good enough to be useable.

Requires backport to v9/v10

How to test manually

webapps

1. Change the value below to true

   webapps/packages/teleport/src/config.ts

   Line 32 in becfad3

   enableDirectorySharing: false, // note to reviewers: should be false in any PRs.

2. run the development server like yarn start-teleport --target=https://ec2-35-171-27-185.compute-1.amazonaws.com/
3. See https://gravitational.slack.com/archives/C02DQ1C2BMW/p1657807702085829 for a username/password combo

What to expect

Go to the top right ... menu and click Share Directory

(X-out of the "uncaught exception" screen if you get it). The directory you select should show up in the File Explorer as a shared drive

The drive should be navigable as well should any folders within it

Trying to open a file should result in a

[JanKaczmarkiewicz]

I initially tried to create a version of listContents which gathered all of the promises in a list and then returned them wrapped in a Promise.all. However with this structure, sharing a directory with a large number of items (like the node_modules directory of this repo after yarn install) would cause the system to freeze up for a while.

I think its a similar problem to the one I encountered way back when working on the display pipeline, namely that piling up a lot of asynchronous calls clogs up the microtask queue and can make performance appear worse, because new regular tasks can't be executed while all the microtasks are processed.

We could use a promise pool eg: https://stackoverflow.com/a/63620273/12860195

[ibeckermayer]

We could use a promise pool eg: https://stackoverflow.com/a/63620273/12860195

@JanKaczmarkiewicz very clever idea and implementation. I tried to copy it in another branch isaiah/sd-list-response-poolPromises. Our system is complicated by the fact that dir.values() gives us an AsyncIterableIterator, so we are most naturally inclined to create an AsyncGenerator. When I attempted to run the code as written, enumerating the shared directory broke the tab.

There may yet be a way to use this, but I'm going to kick the can down the road. Added this discussion thread to the relevant issue: gravitational/teleport#21073