Open joshenlim opened 3 years ago
@ykdojo feel free to clear the rows in our stars
table if you need to for debugging
I'm trying to come up with a hypothesis for why this is happening.
Example: Repo X has 10 stars on Jan 3, 2021. We've already fetched them. On Jan 4, one of the stargazers un-stars the repo. Then the graph should show 9 stars, but when we call the GitHub API... Actually, I'm not sure what it would return exactly then. I'm going to look into a situation like this one - it's probably a good starting point.
lemme know if you need help debugging - we can use our repo.surf repository to test your hypothesis too? Unstar and star ourselves
Yup that's what I was thinking. I think I'll be able to look into this issue tomorrow, too.
I tested some queries with GraphiQL here.
Query:
query {
repository(owner: "supabase", name: "repository.surf") {
stargazers(first: 100, after: null) {
totalCount
edges {
starredAt
cursor
}
}
}
rateLimit {
limit
cost
remaining
resetAt
}
}
Result:
{
"data": {
"repository": {
"stargazers": {
"totalCount": 12,
"edges": [
{
"starredAt": "2020-12-23T16:56:51Z",
"cursor": "Y3Vyc29yOnYyOpIAzg8Rv4A="
},
{
"starredAt": "2020-12-23T17:31:35Z",
"cursor": "Y3Vyc29yOnYyOpIAzg8Ry6g="
},
{
"starredAt": "2020-12-25T02:35:21Z",
"cursor": "Y3Vyc29yOnYyOpIAzg8UlDI="
},
{
"starredAt": "2020-12-25T13:30:35Z",
"cursor": "Y3Vyc29yOnYyOpIAzg8VlIM="
},
{
"starredAt": "2020-12-29T22:04:44Z",
"cursor": "Y3Vyc29yOnYyOpIAzg8ev70="
},
{
"starredAt": "2020-12-30T03:20:07Z",
"cursor": "Y3Vyc29yOnYyOpIAzg8fP0A="
},
{
"starredAt": "2020-12-30T09:38:56Z",
"cursor": "Y3Vyc29yOnYyOpIAzg8gAlw="
},
{
"starredAt": "2020-12-30T14:37:31Z",
"cursor": "Y3Vyc29yOnYyOpIAzg8gkDw="
},
{
"starredAt": "2020-12-30T14:39:00Z",
"cursor": "Y3Vyc29yOnYyOpIAzg8gkRs="
},
{
"starredAt": "2020-12-30T16:24:50Z",
"cursor": "Y3Vyc29yOnYyOpIAzg8gwk8="
},
{
"starredAt": "2021-01-02T03:51:33Z",
"cursor": "Y3Vyc29yOnYyOpIAzg8lbY0="
},
{
"starredAt": "2021-01-04T04:35:04Z",
"cursor": "Y3Vyc29yOnYyOpIAzg8pk_o="
}
]
}
},
"rateLimit": {
"limit": 5000,
"cost": 1,
"remaining": 4996,
"resetAt": "2021-01-09T00:12:55Z"
}
}
}
I unstarred it and ran the same query, and here's the result:
{
"data": {
"repository": {
"stargazers": {
"totalCount": 11,
"edges": [
{
"starredAt": "2020-12-23T16:56:51Z",
"cursor": "Y3Vyc29yOnYyOpIAzg8Rv4A="
},
{
"starredAt": "2020-12-23T17:31:35Z",
"cursor": "Y3Vyc29yOnYyOpIAzg8Ry6g="
},
{
"starredAt": "2020-12-25T13:30:35Z",
"cursor": "Y3Vyc29yOnYyOpIAzg8VlIM="
},
{
"starredAt": "2020-12-29T22:04:44Z",
"cursor": "Y3Vyc29yOnYyOpIAzg8ev70="
},
{
"starredAt": "2020-12-30T03:20:07Z",
"cursor": "Y3Vyc29yOnYyOpIAzg8fP0A="
},
{
"starredAt": "2020-12-30T09:38:56Z",
"cursor": "Y3Vyc29yOnYyOpIAzg8gAlw="
},
{
"starredAt": "2020-12-30T14:37:31Z",
"cursor": "Y3Vyc29yOnYyOpIAzg8gkDw="
},
{
"starredAt": "2020-12-30T14:39:00Z",
"cursor": "Y3Vyc29yOnYyOpIAzg8gkRs="
},
{
"starredAt": "2020-12-30T16:24:50Z",
"cursor": "Y3Vyc29yOnYyOpIAzg8gwk8="
},
{
"starredAt": "2021-01-02T03:51:33Z",
"cursor": "Y3Vyc29yOnYyOpIAzg8lbY0="
},
{
"starredAt": "2021-01-04T04:35:04Z",
"cursor": "Y3Vyc29yOnYyOpIAzg8pk_o="
}
]
}
},
"rateLimit": {
"limit": 5000,
"cost": 1,
"remaining": 4995,
"resetAt": "2021-01-09T00:12:55Z"
}
}
}
Comparing these two, it looks like my starring event was removed (as expected):
{
"starredAt": "2020-12-25T02:35:21Z",
"cursor": "Y3Vyc29yOnYyOpIAzg8UlDI="
},
Then, I tried running the same query with after: "Y3Vyc29yOnYyOpIAzg8UlDI="
.
Result:
{
"data": {
"repository": {
"stargazers": {
"totalCount": 11,
"edges": [
{
"starredAt": "2020-12-25T13:30:35Z",
"cursor": "Y3Vyc29yOnYyOpIAzg8VlIM="
},
{
"starredAt": "2020-12-29T22:04:44Z",
"cursor": "Y3Vyc29yOnYyOpIAzg8ev70="
},
{
"starredAt": "2020-12-30T03:20:07Z",
"cursor": "Y3Vyc29yOnYyOpIAzg8fP0A="
},
{
"starredAt": "2020-12-30T09:38:56Z",
"cursor": "Y3Vyc29yOnYyOpIAzg8gAlw="
},
{
"starredAt": "2020-12-30T14:37:31Z",
"cursor": "Y3Vyc29yOnYyOpIAzg8gkDw="
},
{
"starredAt": "2020-12-30T14:39:00Z",
"cursor": "Y3Vyc29yOnYyOpIAzg8gkRs="
},
{
"starredAt": "2020-12-30T16:24:50Z",
"cursor": "Y3Vyc29yOnYyOpIAzg8gwk8="
},
{
"starredAt": "2021-01-02T03:51:33Z",
"cursor": "Y3Vyc29yOnYyOpIAzg8lbY0="
},
{
"starredAt": "2021-01-04T04:35:04Z",
"cursor": "Y3Vyc29yOnYyOpIAzg8pk_o="
}
]
}
},
"rateLimit": {
"limit": 5000,
"cost": 1,
"remaining": 4994,
"resetAt": "2021-01-09T00:12:55Z"
}
}
}
As you can see, it only has 9 stargazers - everything after my (deleted) starred event.
Then, I starred it again and ran the same query (with after: null
again):
{
"data": {
"repository": {
"stargazers": {
"totalCount": 12,
"edges": [
{
"starredAt": "2020-12-23T16:56:51Z",
"cursor": "Y3Vyc29yOnYyOpIAzg8Rv4A="
},
{
"starredAt": "2020-12-23T17:31:35Z",
"cursor": "Y3Vyc29yOnYyOpIAzg8Ry6g="
},
{
"starredAt": "2020-12-25T13:30:35Z",
"cursor": "Y3Vyc29yOnYyOpIAzg8VlIM="
},
{
"starredAt": "2020-12-29T22:04:44Z",
"cursor": "Y3Vyc29yOnYyOpIAzg8ev70="
},
{
"starredAt": "2020-12-30T03:20:07Z",
"cursor": "Y3Vyc29yOnYyOpIAzg8fP0A="
},
{
"starredAt": "2020-12-30T09:38:56Z",
"cursor": "Y3Vyc29yOnYyOpIAzg8gAlw="
},
{
"starredAt": "2020-12-30T14:37:31Z",
"cursor": "Y3Vyc29yOnYyOpIAzg8gkDw="
},
{
"starredAt": "2020-12-30T14:39:00Z",
"cursor": "Y3Vyc29yOnYyOpIAzg8gkRs="
},
{
"starredAt": "2020-12-30T16:24:50Z",
"cursor": "Y3Vyc29yOnYyOpIAzg8gwk8="
},
{
"starredAt": "2021-01-02T03:51:33Z",
"cursor": "Y3Vyc29yOnYyOpIAzg8lbY0="
},
{
"starredAt": "2021-01-04T04:35:04Z",
"cursor": "Y3Vyc29yOnYyOpIAzg8pk_o="
},
{
"starredAt": "2021-01-08T23:23:01Z",
"cursor": "Y3Vyc29yOnYyOpIAzg82OZA="
}
]
}
},
"rateLimit": {
"limit": 5000,
"cost": 1,
"remaining": 4993,
"resetAt": "2021-01-09T00:12:55Z"
}
}
}
This time, you can see that my starred event moved to the end of this list - kind of as expected.
So, it seems like the only way to make sure that our data is correct is by going through the entire list of stargazers from time to time. We'll probably need some kind of queuing system for this.
With a queuing system, it will look like this:
NOTE: I think we should reserve, say, at least 2000 API calls for on-demand calls.
So we'll reserve ~1000 calls for Supabase, and another ~1000 calls for on-demand calls for other repos. The rest (about 3000) will be available for anything - on-demand calls or queued repos.
I thought of a simpler, easier solution than making a queue system:
We can just allow people to "claim" an org (#33), and then click a button to "refresh" certain repos. I guess this button should be shown when a single repo is selected. We could also show it when multiple repos are selected, in case we want to allow people to refresh multiple repos at once.
Bug report
This one's tricky - it might be that edge case issue that was mentioned. I noticed that the total star count of a repo does not match the latest node in the star history chart as such:
Total star count on the header is 4731 but the latest node in the chart is 4736, checking the github repo directly shows that supabase repo has 4731 stars
Another example under vercel:
Total star count on header is 59687 but latest node in the chart is 59721
One thing's consistent is that our chart is showing slightly higher numbers than the actual