Page MenuHomePhabricator
Differential D3487

[services] Backup - Fix copying big data chunk
ClosedPublic
Actions

Authored by • karol on Mar 22 2022, 1:23 AM.
Tags
None
Referenced Files
Unknown Object (File)
Sun, Nov 24, 9:42 PM
Unknown Object (File)
Sun, Nov 24, 9:42 PM
Unknown Object (File)
Sun, Nov 24, 9:42 PM
Unknown Object (File)
Sun, Nov 24, 9:42 PM
Unknown Object (File)
Sun, Nov 24, 9:42 PM
Unknown Object (File)
Sun, Nov 24, 3:29 PM
Unknown Object (File)
Sat, Nov 16, 10:21 PM
Unknown Object (File)
Fri, Nov 8, 2:24 AM
View All Files
Subscribers
• adrian
ashoat
atul
• benschac
tomek

Details

Reviewers
• max
varun
tomek
• jim
ashoat
marcin
Commits
rCOMM2ca4b0babd9b: [services] Backup - Fix copying big data chunk
Summary

Depends on D3471

Once we receive data as a gRPC field, we don't want to copy it around, especially when it's a data chunk and the size of it may reach up to 4-5MB. Instead, it's always better to have just one place in memory for it and pass the address around.

Test Plan
cd services
yarn run-backup-service

Diff Detail

Repository
rCOMM Comm
Lint
Lint Not Applicable
Unit
Tests Not Applicable

Event Timeline

• karol created this revision.Mar 22 2022, 1:23 AM
Herald added subscribers: • benschac, atul, • adrian and 2 others. · View Herald TranscriptMar 22 2022, 1:23 AM
• karol retitled this revision from [services] Backup - fix copying big data chunk to [draft] [services] Backup - fix copying big data chunk.Mar 22 2022, 1:25 AM
• karol edited the summary of this revision. (Show Details)
• karol edited the test plan for this revision. (Show Details)
• karol added reviewers: • max, varun, tomek, • jim.
• karol added a parent revision: D3471: [services] Blob - Update service implementation.
Harbormaster completed remote builds in B7521: Diff 10586.Mar 22 2022, 1:28 AM
• karol requested review of this revision.Mar 22 2022, 1:28 AM
• karol added a child revision: D3488: [services] Backup - make blob client instances separate for every thread.Mar 22 2022, 2:53 AM
• karol planned changes to this revision.Mar 29 2022, 1:39 AM
• karol updated this revision to Diff 10755.Mar 29 2022, 1:40 AM

rebase

Harbormaster completed remote builds in B7660: Diff 10755.Mar 29 2022, 1:49 AM
• karol planned changes to this revision.Mar 29 2022, 2:58 AM
• karol retitled this revision from [draft] [services] Backup - fix copying big data chunk to [services] Backup - Fix copying big data chunk.
• karol updated this revision to Diff 10761.Mar 29 2022, 4:37 AM

rebase

Harbormaster completed remote builds in B7665: Diff 10761.Mar 29 2022, 4:46 AM
tomek requested changes to this revision.Mar 30 2022, 4:36 AM
tomek added inline comments.
services/backup/docker-server/contents/server/src/Reactors/client/blob/BlobPutClientReactor.h
53–54 ↗(On Diff #10761)

Are you sure this is a good idea? We take this by reference and then move - does it destroy the original data?

This revision now requires changes to proceed.Mar 30 2022, 4:36 AM
• karol requested review of this revision.Apr 1 2022, 4:45 AM
• karol added inline comments.
services/backup/docker-server/contents/server/src/Reactors/client/blob/BlobPutClientReactor.h
53–54 ↗(On Diff #10761)

Yes. I know this looks dangerous, we should use this being aware that the original data will no longer be accessible. I thought it would crash but I ran it and just had an empty string after something like this:

string *str = response.datachunk();
scheduleSendingDataChunk(str);
court << str->size << endl;

Anyway, even though it didn't crash for me, I think we should assume a crash will occur.
For now, I don't see a better way of doing this. I think if we use a unique pointer, we will have the same situation.
The goal here is to prevent copying a big amount of data around.
Even if someone mistakenly uses this value after a function call, they will quickly see a problem there - for sure we should cover this in the tests.
We can add an informative comment there BTW.

What you think?

ashoat requested changes to this revision.Apr 1 2022, 12:09 PM
ashoat added inline comments.
services/backup/docker-server/contents/server/src/Reactors/client/blob/BlobPutClientReactor.h
53–54 ↗(On Diff #10761)

Why would we have the same issue with a unique_ptr? This is basically exactly what unique_ptr is for. When you pass in a unique_ptr to a function, you are saying "this thing is now yours, I no longer can access it" and the compiler will enforce that.

When you pass a reference like you are doing, the caller has no reason to expect that the data will destroyed. When you pass a unique_ptr, the caller knows they cannot access this data ever again.

This revision now requires changes to proceed.Apr 1 2022, 12:09 PM
• karol added inline comments.Apr 4 2022, 1:00 AM
services/backup/docker-server/contents/server/src/Reactors/client/blob/BlobPutClientReactor.h
53–54 ↗(On Diff #10761)

I mean, yes, you're right about the idea behind the unique pointer but you don't seem to be right that the compiler's going to be helpful here. Something like this crashes:

#include <iostream>
#include <memory>

using namespace std;

void f(unique_ptr<string> strp) {
  cout << "f " << *strp << endl;
}

int main()
{
  unique_ptr<string> str = make_unique<string>("asdasd");
  f(move(str));
  cout << "is str nullptr? " << (str == nullptr) << endl; // true
  cout << "out " << *str << endl; // crash

  return 0;
}

So, after a move, unique_ptr holds a nullptr. We can perform a check against the nullptr but if we don't and just try to access the pointer right away and it is null, then it's going to crash.

But I agree that passing a unique pointer here is good because it indicates that the data's not going to be available anymore.

• karol updated this revision to Diff 10982.Apr 4 2022, 3:00 AM

use unique_ptr instead of raw pointers

Harbormaster failed remote builds in B7815: Diff 10982!Apr 4 2022, 3:04 AM
tomek accepted this revision.Apr 4 2022, 4:05 AM
tomek added a comment.Apr 4 2022, 4:07 AM

It looks like there's a build error caused by some other diff

• karol added inline comments.Apr 4 2022, 7:17 AM
services/backup/docker-server/contents/server/src/Reactors/client/blob/BlobPutClientReactor.h
53–54 ↗(On Diff #10761)

Plus, we definitely should avoid using raw pointers as much as possible, so this should be changed.

• karol added a reviewer: marcin.Apr 4 2022, 8:10 AM
ashoat accepted this revision.Apr 5 2022, 1:10 AM

Ah yeah, looks like you're right that the compiler doesn't enforce it. That's unfortunate, but I agree this is still more readable and clear

This revision is now accepted and ready to land.Apr 5 2022, 1:10 AM
• karol updated this revision to Diff 11033.Apr 5 2022, 1:20 AM

rebase

• karol added a comment.Apr 5 2022, 1:20 AM
In D3487#98630, @palys-swm wrote:

It looks like there's a build error caused by some other diff

This is the error I already fixed on master so I'm going to retrigger CI, should be ok!

Harbormaster completed remote builds in B7859: Diff 11033.Apr 5 2022, 1:26 AM
Closed by commit rCOMM2ca4b0babd9b: [services] Backup - Fix copying big data chunk (authored by • karol). · Explain WhyApr 5 2022, 2:14 AM
This revision was automatically updated to reflect the committed changes.
• karol added a commit: rCOMM2ca4b0babd9b: [services] Backup - Fix copying big data chunk.

Revision Contents
Changeset List

PathSize
services/
backup/
docker-server/
contents/
server/
src/
Reactors/
client/
blob/
9 lines
server/
6 lines

Diff 11040

View Options

services/backup/docker-server/contents/server/src/Reactors/client/blob/BlobPutClientReactor.h

Loading...
View Options

services/backup/docker-server/contents/server/src/Reactors/server/CreateNewBackupReactor.h

Loading...