Skip to content
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

fix: virtualize pool #4706

Closed
wants to merge 1 commit into from
Closed

fix: virtualize pool #4706

wants to merge 1 commit into from

Conversation

erights
Copy link
Member

@erights erights commented Mar 1, 2022

Virtualize pool. Unlike #4707, this one does not virtualize vpool (the single pool) because it seems that the vpool need not be made persistent.

See #4707 (comment)

Note to reviewers: I may still be confused. I suspect that @Chris-Hibbert and I need to pair program this before it is good.

@erights erights self-assigned this Mar 1, 2022
@erights erights changed the base branch from master to markm-define-heap-kind March 1, 2022 07:03
@Chris-Hibbert
Copy link
Contributor

I'm expecting to see a test demonstrating that objects can be defined, saved, flushed from memory, and restored with behavior identical to the original. Is there a reason that's not appropriate here?

@erights erights marked this pull request as draft March 21, 2022 19:29
@erights erights force-pushed the markm-multi-facet-pool branch from 3e0e502 to 72cfb21 Compare March 21, 2022 20:06
@erights erights changed the base branch from markm-define-heap-kind to master March 21, 2022 20:07
@erights erights marked this pull request as ready for review March 21, 2022 20:15
@erights erights marked this pull request as draft March 21, 2022 20:16
@erights erights force-pushed the markm-multi-facet-pool branch 2 times, most recently from fed23b2 to 7164f83 Compare March 21, 2022 20:40
@erights erights changed the base branch from master to markm-virtualize-ERTP March 21, 2022 20:42
@erights erights changed the base branch from markm-virtualize-ERTP to master March 21, 2022 20:45
@erights erights changed the base branch from master to markm-virtualize-ERTP March 21, 2022 20:45
@erights erights changed the base branch from markm-virtualize-ERTP to master March 21, 2022 20:47
@erights erights changed the base branch from master to markm-virtualize-ERTP March 21, 2022 20:47
@erights erights force-pushed the markm-multi-facet-pool branch from 1592e43 to 2cfe4a3 Compare March 21, 2022 22:16
@erights
Copy link
Member Author

erights commented Mar 21, 2022

At #4706 (comment) @Chris-Hibbert asks:

I'm expecting to see a test demonstrating that objects can be defined, saved, flushed from memory, and restored with behavior identical to the original. Is there a reason that's not appropriate here?

@FUDCo , should this PR test that? If so, how? Thanks.

@erights erights marked this pull request as ready for review March 21, 2022 22:33
@erights erights marked this pull request as draft March 21, 2022 23:01
@erights
Copy link
Member Author

erights commented Mar 21, 2022

Converting back to draft because it is only a step towards #4707 . I'll let you know when #4707 is ready for review.

@erights erights force-pushed the markm-virtualize-ERTP branch 2 times, most recently from a4fb604 to a68c837 Compare March 22, 2022 19:53
Base automatically changed from markm-virtualize-ERTP to master March 22, 2022 20:47
@erights erights force-pushed the markm-multi-facet-pool branch from 2cfe4a3 to 8d7d290 Compare March 23, 2022 03:35
@erights erights marked this pull request as ready for review March 23, 2022 03:55
@erights
Copy link
Member Author

erights commented Mar 23, 2022

Given @Chris-Hibbert 's comments at #4707 (comment) I put #4707 into draft status. Instead, it seems this PR #4706 is closer to what we want, so I'm declaring #4706 Ready to Review. Everyone, please review this one instead.

@erights erights changed the title fix: virtually virtualize pool fix: virtualize pool Mar 23, 2022
@erights erights marked this pull request as draft March 23, 2022 04:11
@erights
Copy link
Member Author

erights commented Mar 23, 2022

Changed my mind again. Back to Draft until @Chris-Hibbert and I pair on this and it becomes ready. No need for others to review before then.

@warner
Copy link
Member

warner commented Mar 23, 2022

At #4706 (comment) @Chris-Hibbert asks:

I'm expecting to see a test demonstrating that objects can be defined, saved, flushed from memory, and restored with behavior identical to the original. Is there a reason that's not appropriate here?

@FUDCo , should this PR test that? If so, how? Thanks.

We don't have a way to exercise that until #1848 / #4382 is done and we can upgrade vats. We've got test coverage on durable objects/collections in general via packages/SwingSet/ .. that will have to be enough until we get upgrade ready for use.

Although.. I suppose if you had a swingset-based test in which the virtual/durable collections were used to create some objects in vat A, then you carefully drop all references to them from RAM (leaving only exports and/or refs from virtualized data), then you wait a crank or two, then send a message back into vat A that references those objects.. that would exercise the code that rebuilds the objects from saved data.

The tricky part is knowing that the original objects really got GCed, which userspace doesn't get to see. You could add some temporary debug messages to liveslots.js while testing the test, to convince yourself that GC is really happening.

Running the workers under XS (config.defaultManagerType = 'xs-worker') is a good idea, that makes GC far more predictable.

@Chris-Hibbert
Copy link
Contributor

We don't have a way to exercise that until #1848 / #4382 is done and we can upgrade vats. We've got test coverage on durable objects/collections in general via packages/SwingSet/ .. that will have to be enough until we get upgrade ready for use.

I was more interested in virtualization than durability. Can we page objects out and back in at this point?

liquidityValueIn,
),
);
const makePool = defineKind(
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

is the best documentation on defineKind the comment on defineKindInternal in virtualObjectManager.js? The parameter types don't seem to be declared in a way that's visible to VSCode and WebStorm, so I always have to find the docs before I can tell what arguments should be provided in what order.

Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Copy link
Contributor

@Chris-Hibbert Chris-Hibbert left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

This looks fine and straightforward. I asked one question about how makeKind works.

Do we have the ability to demonstrate that virtualization can handle paging objects out and back in again?

I've asked before and it seems like the responses have assumed I was asking about durability and revival of the contract rather than the pool objects.

);
const makePool = defineKind(
'pool',
(liquidityZcfMint, poolSeat, secondaryBrand) => {
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I'm trying to understand defineKind. How does it know what parameters to provide to the init function? (i.e. If the pool doesn't declare them somewhere, what rules produce the parameters?)

Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

defineKind doesn't "know" the parameters to init. The maker function calls init with whatever parameters it was given.


/** @type {XYKPool} */
const pool = {
getLiquiditySupply: () => state.liqTokenSupply,
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

why is liqTokenSupply access through state, but liquidityIsuser, poolSeat, updater, etc. (line 126) are extracted and accessed as top-level values?

@FUDCo
Copy link
Contributor

FUDCo commented Mar 23, 2022

At #4706 (comment) @Chris-Hibbert asks:

I'm expecting to see a test demonstrating that objects can be defined, saved, flushed from memory, and restored with behavior identical to the original. Is there a reason that's not appropriate here?

@FUDCo , should this PR test that? If so, how? Thanks.

I don't think it should have to be the responsibility of users of the virtual object system to write tests to verify that the virtual object system works as advertised. The virtual objects implementation itself does have tests for that though.

@Chris-Hibbert
Copy link
Contributor

I don't think it should have to be the responsibility of users of the virtual object system to write tests to verify that the virtual object system works as advertised. The virtual objects implementation itself does have tests for that though.

I'm willing to accept that it works as advertised. As a reviewer of a conversion, I'm trying to gain confidence that the right state has been captured. I'd feel a lot more confident if I saw that objects that had been paged out and in behaved the same as before. Without that I have to have confidence that I've identified all the state that is in use, and that I understand how the parts of makeKind interact in great detail.

@FUDCo
Copy link
Contributor

FUDCo commented Mar 23, 2022

I don't think it should have to be the responsibility of users of the virtual object system to write tests to verify that the virtual object system works as advertised. The virtual objects implementation itself does have tests for that though.

I'm willing to accept that it works as advertised. As a reviewer of a conversion, I'm trying to gain confidence that the right state has been captured. I'd feel a lot more confident if I saw that objects that had been paged out and in behaved the same as before. Without that I have to have confidence that I've identified all the state that is in use, and that I understand how the parts of makeKind interact in great detail

Ah, I misunderstood the intent of what you were trying to get at. That makes a lot of sense and seems like a thing that's genuinely important to verify.

@erights
Copy link
Member Author

erights commented Mar 23, 2022

@FUDCo

Ah, I misunderstood the intent of what you were trying to get at. That makes a lot of sense and seems like a thing that's genuinely important to verify.

Is there a way to verify that? Should this PR do so? How?

@turadg
Copy link
Member

turadg commented Apr 22, 2022

Done in #5187

@turadg turadg closed this Apr 22, 2022
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

Successfully merging this pull request may close these issues.

5 participants