-
Notifications
You must be signed in to change notification settings - Fork 5
Commit
This commit does not belong to any branch on this repository, and may belong to a fork outside of the repository.
server: fix ordering of explore query, simplify using backend state
- Loading branch information
Showing
1 changed file
with
82 additions
and
113 deletions.
There are no files selected for viewing
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| Original file line number | Diff line number | Diff line change |
|---|---|---|
| @@ -1,136 +1,105 @@ | ||
| import { Request, Response } from 'express'; | ||
| import { prisma } from '../../client.js'; | ||
| import { logger as parentLogger } from '../../logger.js'; | ||
| import { _getDpidForTxIds, _getIndexedResearchObjects, getIndexedResearchObjects } from '../../theGraph.js'; | ||
| import { decodeBase64UrlSafeToHex, hexToCid } from '../../utils.js'; | ||
|
|
||
| const logger = parentLogger.child({ | ||
| module: 'NODE::exploreController', | ||
| }); | ||
|
|
||
| /** | ||
| * Traverse all published nodes. This should ideally be done by the resolver, | ||
| * but while we need to union over legacy and new dPID registries it's much | ||
| * easier to use the DB state to figure out where to look for histories. | ||
| * | ||
| * Tradeoffs: | ||
| * - Using the nodes DB to find nodes | ||
| * - If no dpidAlias, falling back to querying the dpid SG by tx id | ||
| * - Have to be backward compatible with the old resolver format | ||
| * - Order of node selection depends on node.updatedAt instead of the actual pub | ||
| * - Unecessarily digs through historical versions in the index | ||
| In case it's needed to debug/tune on dev/prod instances, this is the SQL query | ||
| that's translated to prisma below: | ||
| - Joins nodes with nodeVersions if there is a non-null anchor/transaction hash | ||
| - This join is grouped by node ID, and sorted by version creation | ||
| - Selects one row per node ID, which yields the top (latest) one | ||
| - From this last-publish CTE, we sort on the creation timestamp and take our page | ||
| WITH latest_versions AS ( | ||
| SELECT DISTINCT ON (n.id) | ||
| n.id, | ||
| n.uuid, | ||
| n."dpidAlias", | ||
| nv."manifestUrl" as "recentCid", | ||
| nv."createdAt" as "latestPublish", | ||
| nv."commitId", | ||
| nv."transactionId" | ||
| FROM | ||
| "Node" n | ||
| INNER JOIN "NodeVersion" nv ON n.id = nv."nodeId" | ||
| WHERE | ||
| nv."transactionId" IS NOT NULL OR nv."commitId" IS NOT NULL | ||
| ORDER BY | ||
| n.id, | ||
| nv."createdAt" DESC, | ||
| nv.id DESC -- determinism in case of same anchor time on fast publishes | ||
| ) | ||
| SELECT | ||
| "latestPublish", | ||
| uuid, | ||
| "dpidAlias", | ||
| "recentCid", | ||
| "commitId", | ||
| "transactionId" | ||
| FROM latest_versions | ||
| ORDER BY "latestPublish" DESC | ||
| LIMIT $size | ||
| OFFSET ($page - 1) * $size | ||
| The corresponding prisma query build is a bit backward as we start from | ||
| nodeVersion and use nested select instead of the join, | ||
| */ | ||
|
|
||
| /** | ||
| * Get the latest publish information for the most recently updated nodes, | ||
| */ | ||
| export const explore = async (req: Request, res: Response) => { | ||
| const page: number = req.query.page ? parseInt(req.query.page as string) : 1; | ||
| const size: number = req.query.size ? parseInt(req.query.size as string) : 10; | ||
|
|
||
| logger.info({ page, size }); | ||
| if (size > 10) { | ||
| logger.warn({ size }, "explore queries get progressively slower for large sets, under 10 is best"); | ||
| } | ||
|
|
||
| let nodes = await prisma.node.findMany({ | ||
| select: { | ||
| uuid: true, | ||
| dpidAlias: true, | ||
| updatedAt: true, | ||
| versions: { | ||
| select: { | ||
| transactionId: true, | ||
| commitId: true, | ||
| }, | ||
| where: { | ||
| OR: [ | ||
| { transactionId: { not: null }}, | ||
| { commitId: { not: null }}, | ||
| ], | ||
| }, | ||
| orderBy: { | ||
| createdAt: "desc", | ||
| }, | ||
| take: 1, | ||
| try { | ||
| let freshlyPublishedVersions = await prisma.nodeVersion.findMany({ | ||
| where: { | ||
| OR: [ | ||
| { transactionId: { not: null } }, | ||
| { commitId: { not: null } } | ||
| ] | ||
| }, | ||
| }, | ||
| where: { | ||
| versions: { | ||
| // true if there were any versions with publish event IDs | ||
| some: { | ||
| OR: [ | ||
| { transactionId: { not: null }}, | ||
| { commitId: { not: null }}, | ||
| ], | ||
| }, | ||
| orderBy: [ | ||
| { createdAt: "desc" }, | ||
| { id: "desc" }, | ||
| ], | ||
| select: { | ||
| manifestUrl: true, | ||
| createdAt: true, | ||
| commitId: true, | ||
| transactionId: true, | ||
| node: { | ||
| select: { | ||
| uuid: true, | ||
| dpidAlias: true, | ||
| } | ||
| } | ||
| }, | ||
| }, | ||
| orderBy: { | ||
| updatedAt: "desc", | ||
| // Ideal: sort on publish date, smt like this: | ||
| // versions: { createdAt: "desc" } | ||
| }, | ||
| take: size, | ||
| skip: (page - 1) * size, | ||
| }); | ||
|
|
||
| /* nodes-web (current) requirements: | ||
| - uses id only as item keys, can be anything unique | ||
| - needs dpid field, used onClick to resolve node | ||
| - needs recentCid for: title, authors | ||
| - uses date from versions[-1].time | ||
| - title from manifest | ||
| distinct: ['nodeId'], | ||
| take: size, | ||
| skip: (page - 1) * size, | ||
| }); | ||
|
|
||
| Could probably figure this stuff out from the database state though | ||
| to make this mess simpler and faster, if that works for the webapp. | ||
| const flattened = freshlyPublishedVersions.map(v => ({ | ||
| id: v.node.uuid, | ||
| // Chop off the milliseconds to match epoch format | ||
| time: (v.createdAt.valueOf()).toString().slice(0, -3), | ||
| recentCid: v.manifestUrl, | ||
| dpid: v.node.dpidAlias, | ||
| commitId: v.commitId, | ||
| transactionId: v.transactionId, | ||
| })); | ||
|
|
||
| TODO: The time it takes to get the entire history for each stream increases | ||
| with the number of objects being checked, as we only use the timestamp | ||
| this is unnecessary. | ||
| */ | ||
| try { | ||
| const indexedObjectsPromise = getIndexedResearchObjects( | ||
| nodes.map(n => n.uuid) | ||
| ); | ||
|
|
||
| const legacyDpidsPromise = _getDpidForTxIds(nodes | ||
| .map(n => n.versions[0].transactionId) | ||
| .filter(Boolean) | ||
| ); | ||
|
|
||
| const [indexedObjects, legacyDpids] = await Promise.all( | ||
| [indexedObjectsPromise, legacyDpidsPromise] | ||
| ); | ||
|
|
||
| if (indexedObjects.researchObjects.length !== nodes.length) { | ||
| logger.warn({ indexedObjects, nodes }, "Indexers and database ") | ||
| }; | ||
|
|
||
| const txToLegacyDpid = legacyDpids.reduce( | ||
| (acc, ld) => ({ ...acc, [ld.transactionHash]: parseInt(ld.entryId) }), | ||
| ({} as Record<string, string>), | ||
| ); | ||
|
|
||
| const withDpid = indexedObjects.researchObjects | ||
| .toSorted((o1, o2) => parseInt(o1.versions[0].time) - parseInt(o2.versions[0].time)) | ||
| .map(indexObject => { | ||
| const node = nodes.find( | ||
| n => `0x${decodeBase64UrlSafeToHex(n.uuid)}` === indexObject.id | ||
| ); | ||
| return { | ||
| id: indexObject.id, | ||
| dpid: node.dpidAlias ?? txToLegacyDpid[node.versions[0].transactionId] ?? null, | ||
| recentCid: hexToCid(indexObject.recentCid), | ||
| researchObject: { | ||
| ...indexObject, | ||
| versions: indexObject.versions.map(v => ({ | ||
| ...v, | ||
| cid: hexToCid(v.cid), | ||
| time: parseInt(v.time), | ||
| })) | ||
| } | ||
| } | ||
| }); | ||
| return res.send(withDpid); | ||
| return res.send(flattened); | ||
| } catch (e) { | ||
| logger.error({ e, nodes }, "explore query failed to complete"); | ||
| logger.error({ e, page, size }, "explore query failed to complete"); | ||
| return res.status(500).send(); | ||
| } | ||
| }; | ||
| }; |