Send the message catalogue by reference, not inside every document
The root layout handed the whole catalogue to the client provider as a prop, so React inlined it into the RSC payload of every document: on /leads/calling 2.96 ...
Read a drip's lead by its primary key
DripContacts.lead/stateOfLead wrote "l.uid = ? AND l.org_id = ?", which YugabyteDB's heuristic planner serves from ix_leads_org_id, walking the whole organizati...
Answer an outcome's status write before its follow-ups
Declares LeadStatusFollowUps (shared-lib 5b273d8e): the lead's last contact and pipeline stage after a status write now run on two background threads (bounded q...
Answer a status write before its follow-ups, and read lead rows by their own key
Call Floor outcome press, profiled with JFR on a SecurityService pod (lab org, 40 presses): the /stage write took 160-360 ms, median ~220 ms. About 90 ms of it ...
Paint the next lead first, read ahead during a call, and save the outcome behind the next lead
First paint: the list is asked for at once, beside the member's rights, and its top (limit=6) comes first so the next lead paints before the rest of the list; a...
Check a page of leads in one read, and read a lead's loans by index seeks
OutreachCheck.checkPage answers a person's check for a page of leads with the same rules and the same decision code as check(): the leads' rows and open points,...
Read the pipeline's borrowers beside its rows, not after them
Each of the reader's two queries takes about a second for a large lender, and they ran back to back. The borrower read now runs on its own connection while the ...
My Tasks shows each source as it lands, and asks for far less
Lead names come from one batch read (/leads/names) instead of a full lead read per task. Deals are read only when the Deal filter is opened, four leads at a tim...
Several leads' names in one read, on the lead page's own rule
GET /api/security/leads/names?ids=: one findAllById for up to 100 leads, each through LeadViewGate, VIEW_LEADS to ask. My Tasks read a full lead per task (200-6...
Background work uses three engine slots per pod and is sent as low priority
One background slot per pod left three of the engine's four streams idle during a dictionary sync. Background callers (a wait above the 100-second default: the ...
Four streams of four threads, and a lane that keeps one free for chat
Measured on the nodes' CPU class with the dictionary sync's own traffic (15-key chunks): one stream of eight threads did 1.5 keys/s, two of eight 1.8, four of f...
Beam 4 only for labels, beam 2 up to 64 tokens
Cost is tokens x beam. Measured in the cluster with the old engines still running: one stream 77 chars/s and two 115 chars/s on short strings at beam 4. Sentenc...
The mortgage pipeline loads in a handful of queries, and a deal can be renamed and take a second loan
The pipeline's two reads were N+1 on the round trip: - GET /mortgage-apps loaded plain AppMortgage rows, and each one then fired a SELECT per inverse @OneToO...
Index member_notifications by (source_type, source_id) for withdrawal across members (IoT close-out F15)
The withdraw-by-source read (findBySourceTypeAndSourceId) and the IoT alarm's withdraw-by-prefix (StartingWith -> LIKE 'x%') had only the member-led index, so b...
The read pointer moves while the authors are checked
chat_read made three MediaService calls one after another (the page, the session for the inside-content check, then the read pointer). The pointer is now moved ...
An SI shows "typing…" through an internal endpoint, and a message's badges are recounted off the sender's request
- POST /api/media/internal/chat/typing {memberId, conversationId, typing} (KamoAI internal secret, InteractionInternalAuth; under /api/media/internal, which the...
VISIBLE as one id IN (… UNION ALL …), so every visible note read is index scans bounded by a key, not a scan of every note (KamoCollab CP05 Task 15)
Yugabyte, with bitmap scans off, answers the three ORed branches across different columns of notes with a sequential scan (EXPLAIN in production, the pre-CP05 t...
A turn-only fan-out reads no roster, ticket id or conversation title
SP98-D-6. ChatFanoutService.prepare resolved the roster, the support ticket's id and the conversation's title for every request, including the turn-only ones (a...
A template icon is drawn through Batik once, not on every send
InlineImageTransport memoised a drawing only within one message, so every send that wore the compose template decoded and transcoded the same handful of social ...
Six Whisper workers and six requested CPUs, so a batch chunk never queues a realtime turn
The acceptance run on k3m1 (4 realtime streams + 1 batch) measured p95 17.9 s: the batch held one of the 4 workers, so one turn waited behind a whole turn. Six ...
Maintain commit_logs.translation_count at write time
DDL: **************** (column + partial-friendly index, applied to production already) and **************** (one-time catch-up, already run — 19,373 rows, all l...
Use the lean LeadVendorProduct reads, not the 8-way eager findById
Caller half of the kamo-shared-library fix: switches every findById() call on LeadVendorProductRepository in securityservice to whichever lean method matches wh...
CommitLog carries a maintained translation_count
Adds CommitLog.translationCount and **************** the library side of replacing the changelog translation-backfill sweep's GROUP BY .. HAVING scan with a pla...
Member_rights_applied writes only the difference, and reads roles in bulk
Two related fixes to MemberRightsAppliedService, both in the same file: 1. **************** (via recalcChunk) and **************** unconditionally deleted a...
Stop eager-fetching LeadVendorProduct's org chain for a single-id read
findById() on LeadVendorProduct pulls its default-EAGER vendor, which itself eager-fetches its own org AND its market (which eager-fetches ITS org again) — and ...
Read the public changelog's commit counts from a ledger, not COUNT(*)
pg_stat_statements: `SELECT COUNT(c) FROM CommitLog c WHERE c.project IN (:~33 public projects) [AND c.commitTypeId = :t]` ran ~140K times a day at 100-200 ms e...
Fold and recount the public changelog's commit-log ledger
securityservice's create_commit_log_ledger.sql (applied by hand as the owner) adds commit_log_counts / commit_log_count_deltas: a total per (project, commit typ...
Stop asking three hot queries one row at a time
Three queries pg_stat_statements showed as kbservice's top production DB cost, all from asking in a loop (or with a plan the legacy YugabyteDB planner picked ba...
Batch two translation lookups the planner was doing badly per row
kbservice's public KB endpoints and its translation-retry sweep both asked KbArticleTranslationRepository one row at a time. Production's pg_stat_statements sho...
An entry page reads three rows, not the whole public changelog
pg_stat_statements put the public changelog's single-entry reads at the top of the database by a distance: the slug lookup and the two neighbour lookups ran ~90...
Calendar reminder sends move off the shared scheduler thread
Spring's default single scheduler thread is shared by every @Scheduled method in this service — the 1-second campaign pulse, the drip engine's tick, and Reminde...
Bound and stream the byte-heavy imaging paths
/download loaded the whole file into a byte[] through ImageService.downloadDocument before serving it; /bulk-download built the whole ZIP (or merged PDF) as one...
Don't load content_json to build the public tree index, and stop loading it twice
buildPublicTreeIndex builds a small structural index (parent/child, slug path, ancestors) from every public article in the org, but it was loading full KbArticl...
Manage-Credits and Leads-Available read the lead ledgers once instead of counting per row
/leads/credits (GET /api/security/credits/manage) counted the lead pool once per allotment row — ~8 s apiece against one org's 796k-lead pool — then once more g...
Leads available and credit balances read a ledger instead of counting rows
"Leads available" was COUNT(*) over the assignable pool, and one org's pool holds ~796k leads: ~7.8 s per (market, product) count and ~11 s for the grouped one ...
The dock moves on the compositor, and a window that moves no longer re-renders the tool inside it
Everything about a popup tool window that MOVES - re-packing the row, trading places with a strip click, stepping out of a row behind, a drag, a seam pull, taki...
Check pool eligibility with an unordered sample, not a full-pool sort
hasEligibleAssignableLead only needs to know whether any pool lead is one the member may take, but it read a page of findAssignablePool, whose ORDER BY assigned...
Create intake leads concurrently and claim larger batches
The job created one lead at a time from a 500-row claim, ~230 leads/min at best, which turns a large vendor list into days of backlog. It now claims 2000 payloa...
The clock drains on the compositor, not on a timer
The felt's turn clock ran off a 200ms setInterval writing a fraction into React. That is an honest way to build a bar that must track a server deadline, and it ...
Stop rebuilding the wallet and the guest list on every check
`viewFor` fetched the recipient's Cheddar balance and queried the whole organization's member list — both per RECIPIENT, on every published frame. At a full tab...
Stop every scene re-rasterising its sprite bank on resize
An audit of all eighteen scenes found the same defect in twelve of them, and it is the shape that degrades a tab over a minute or two and then stalls it in garb...
Narrow the window a DDL leaves behind for stale backends
Applying DDL bumps the catalog version, and backends still holding the old one fail their next statement with SQLState 40001 until they are recycled. The last o...
The profile endpoint stops loading the whole board
Its own comment said it existed so a page could paint the employer "without pulling every advert down with it", and the line under that comment called board() —...
Stop the inline image toolbar re-rendering on every scroll frame
measure() runs on each scroll and resize event and built a fresh state object every time, so the toolbar re-rendered whether or not anything about the image had...
Index a message in one statement instead of a select-then-write
The search index row's key is assigned rather than generated, so handing a built entity to save() was a merge: Spring Data SELECTed the whole row — body text in...
Stop repeating the ancestor chain on every node of a tree
A node inside a tree already knows where it is -- its position is the tree. Sending the ancestor chain on each one meant every article carried two ancestor refs...
A conversation outlives the window showing it
Opening a chat window cost seven requests before a message rendered — history, members, receipts, mark-read, the unread snapshot, the counterpart's avatar and t...
Serve a reopened window only what it missed, and audit a page in one write
Two costs, both paid on every single open of a chat window. ?since on GET /sessions/{guid}/messages. A window being reopened wants the messages after the one i...
One session snapshot for the tab, not one per hook instance
useUserInfo held its session in three useState cells. It has 187 call sites and roughly 27 of them are mounted for the whole session — the nav, every shell prov...
Like what you see shipping?
All of it arrives in your workspace on its own. Start on the free plan and read this page again in a month.