Skip to content

feat(workflow): match search and listings against the copy on show - #7857

Closed
yangzhang75 wants to merge 4 commits into
apache:mainfrom
yangzhang75:pin/4-search
Closed

feat(workflow): match search and listings against the copy on show#7857
yangzhang75 wants to merge 4 commits into
apache:mainfrom
yangzhang75:pin/4-search

Conversation

@yangzhang75

Copy link
Copy Markdown
Contributor

Part of #7828. Stacked on #7856 — the review here is the last commit, feat(workflow): match search and listings against the copy on show.

Search and the listings it feeds read the author's live columns. For a pinned workflow that is the one copy the public cannot open: a draft would turn up in a public search under a title nobody has seen, and the card would advertise a name the detail page does not show.

One filter, applied to the copy the caller may see

onVisibleCopy builds each filter twice — over the live columns for rows the caller was granted access to, over the frozen ones for rows they reach only because the workflow is public — and ORs the two. A disjunction over bare columns rather than a CASE, so each side stays eligible for its own fulltext index. Unpinned public rows fall back to the live columns, so a following workflow searches exactly as it does today.

This covers both halves of what public search indexes: the words in a workflow, and the operators in it.

Two more things listings carry

From the same query, no second round trip:

  • the frozen name and description, shown to a viewer without granted access, so a listing agrees with what opening it shows;
  • whether the copy on show is behind the author's working copy — an aggregate, which is what a hub card needs to know to open the copy it is advertising.

The hub's own listing query gets the same treatment. It is the public shelf, so everything on it is listed as the public sees it — the author included, who is looking at the shelf and not at their own dashboard.

constructWhereClause takes includePublic for this; the dataset and project builders accept and ignore it.

Tests

+27 backend cases. Search: a draft is unfindable by a keyword or an operator that exists only in it, while the author still finds their own; a pinned workflow matches its frozen title and not its live one, and an unpinned one matches its live title. Listings: the frozen values shown to a stranger and the live ones to the author and to a collaborator; drift reported only for a workflow that actually has a frozen copy. Plus the record-mapping cases, including that an unknown access answer counts as public — the reverse is the leak.

yangzhang75 and others added 4 commits August 22, 2026 14:36
A public workflow follows the author's latest content today: every save reaches
the Hub immediately. Pinning a version as the public copy needs somewhere to keep
that copy, which is what these columns are.

`is_public` stays the on/off switch. `published_content` is the pin: NULL means
the workflow follows the author's latest, which is what every workflow does today,
so the migration changes nothing anyone can see. `published_name` and
`published_description` travel with it because a pin has to hold everything on
show, and `workflow_version` stores no metadata at all -- only content deltas.
`published_version_id` names the version row holding that copy, so the revision
panel can mark it and the author can restore it.

The copy is materialized rather than replayed from `workflow_version` because
those rows are reverse JSON-Patch deltas: serving a pinned workflow would mean
folding every newer patch back from the author's current content on each public
read, and a computed value is something the fulltext index cannot cover.

A CHECK constraint makes "private but pinned" unrepresentable, and a PGroonga
index mirrors the latest-content one so public search can match the frozen copy.

Adding columns changes the arity of the generated positional constructor, so the
three copy-producing paths (clone, duplicate, restore-a-version) now build their
POJO with setters -- which is also what stops a later column from silently
shifting a null into the wrong field.

Part of apache#7828.
A public workflow follows the author's latest content, as publishing has
always done. This adds the other state: the author pins the version they
have now, and the public copy stops moving until they pin again.

`is_public` stays the on/off switch; `published_content` is the pin, NULL
while following. `WorkflowPublishService` owns the two states, and three
endpoints expose them: POST and DELETE `/workflow/pin/{wid}` to pin and
unpin, GET `/workflow/publish-status/{wid}` for what the author is shown.
Publishing and unpublishing move through the same service, so unpublishing
drops the pin rather than leaving a private workflow carrying one.

Two paths are narrowed so a pin can hold. A save wrote the whole row back,
so a publish landing while a save was in flight was silently rolled back,
and a request body could set the publish columns itself; saves now write
only name, description and content. Creating a workflow clears the publish
columns for the same reason.

Nothing reads the pinned copy yet: every workflow is in the following
state it is in today, and nothing on screen changes.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
With a version pinned, a workflow has two copies: the author's working
copy and the frozen one on public show. This routes every read that
serves a viewer without granted access through the frozen copy, and
freezes the name and description with the graph.

`WorkflowPublishService.publicCopyOf` returns the three fields as a
group, so a surface cannot pick up the published graph under a title the
author has not published; `WorkflowAccessResource.hasGrantedAccess` is
the seam that decides which copy a caller gets. Granted access -- owner,
shared, project member -- keeps tracking the author's latest, because
sharing is not publishing.

Name and description freeze because they are as public as the graph: if
only the graph froze, a report about a title could be answered by editing
the title while the pinned copy still advertised it.

Routed through it: opening a workflow, the hub's read, Clone, Duplicate,
`/workflow_name`, `/workflow_description` and the size a listing shows.
A workflow that follows the author's latest -- every workflow today --
is served exactly what it is served now.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Search and the listings it feeds were reading the author's live columns,
which for a pinned workflow is the one copy the public cannot open. A
draft would turn up in a public search under a title nobody has seen, and
the card would advertise a name the detail page does not show.

Each filter is now applied to whichever copy the caller may see:
`onVisibleCopy` builds the same filter twice -- over the live columns for
rows the caller was granted access to, over the frozen ones for rows they
reach only because the workflow is public -- and ORs the two. A
disjunction over bare columns rather than a CASE, so each side stays
eligible for its own fulltext index. Unpinned public rows fall back to
the live columns, so a following workflow searches exactly as it does now.

Listings carry two more things from the same query: the frozen name and
description to show a viewer without granted access, and whether the copy
on show is behind the author's working copy. `constructWhereClause` takes
`includePublic` for this; the other builders accept and ignore it.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
@github-actions

Copy link
Copy Markdown
Contributor

Automated Reviewer Suggestions

Based on the git blame history of the changed files, we recommend the following reviewers:

  • Contributors with relevant context: @aglinxinyuan, @tanishqgandhi1908, @Mrudhulraj
    You can notify them by mentioning @aglinxinyuan, @tanishqgandhi1908, @Mrudhulraj in a comment.

@codecov-commenter

codecov-commenter commented Aug 23, 2026

Copy link
Copy Markdown

Codecov Report

❌ Patch coverage is 91.66667% with 17 lines in your changes missing coverage. Please review.
✅ Project coverage is 91.89%. Comparing base (b7c33b0) to head (e015b5a).
⚠️ Report is 3 commits behind head on main.

Files with missing lines Patch % Lines
...shboard/user/workflow/WorkflowPublishService.scala 81.33% 4 Missing and 10 partials ⚠️
...ra/web/resource/dashboard/SearchQueryBuilder.scala 50.00% 1 Missing ⚠️
...esource/dashboard/WorkflowSearchQueryBuilder.scala 97.87% 1 Missing ⚠️
...rce/dashboard/user/workflow/WorkflowResource.scala 98.38% 0 Missing and 1 partial ⚠️
Additional details and impacted files
@@             Coverage Diff              @@
##               main    #7857      +/-   ##
============================================
+ Coverage     91.87%   91.89%   +0.02%     
- Complexity     4510     4516       +6     
============================================
  Files          1173     1174       +1     
  Lines         47350    47501     +151     
  Branches       5306     5334      +28     
============================================
+ Hits          43502    43652     +150     
+ Misses         2204     2195       -9     
- Partials       1644     1654      +10     
Flag Coverage Δ *Carryforward flag
access-control-service 81.00% <ø> (ø)
agent-service 98.62% <ø> (ø) Carriedforward from 8f56d11
amber 88.47% <91.66%> (+0.09%) ⬆️
computing-unit-managing-service 73.67% <ø> (ø)
config-service 86.73% <ø> (ø)
file-service 75.74% <ø> (ø)
frontend 93.79% <ø> (ø) Carriedforward from 8f56d11
notebook-migration-service 79.13% <ø> (ø)
pyamber 97.57% <ø> (ø) Carriedforward from 8f56d11
workflow-compiling-service 77.19% <ø> (ø)

*This pull request uses carry forward flags. Click here to find out more.

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:
  • 📦 JS Bundle Analysis: Save yourself from yourself by tracking and limiting bundle sizes in JS merges.

@github-actions

Copy link
Copy Markdown
Contributor

⚠️ Benchmark changes need a look

🟢 0 better · 🔴 15 worse · ⚪ 0 noise (<±5%) · 0 without baseline

CI benchmark results are noisy; treat <±5% as noise unless repeated.

Dashboard · Run

config throughput MB/s latency max Δ latest / 7d
🔴 bs=10 sw=10 sl=64 419 0.256 23,252/31,412/31,412 us 🔴 +158.7% / 🔴 +118.4%
🔴 bs=100 sw=10 sl=64 946 0.577 105,361/130,178/130,178 us 🔴 +73.3% / 🔴 +30.8%
🔴 bs=1000 sw=10 sl=64 1,103 0.673 907,832/943,462/943,462 us 🔴 +32.3% / ⚪ within ±5%
Baseline details

Latest main b7c33b0 from 2026-08-22T12:41:18.215Z

config metric PR latest main 7d avg Δ latest Δ 7d
bs=10 sw=10 sl=64 throughput 419 tuples/sec 1,070 tuples/sec 833.79 tuples/sec -60.8% -49.7%
bs=10 sw=10 sl=64 MB/s 0.256 MB/s 0.653 MB/s 0.509 MB/s -60.8% -49.7%
bs=10 sw=10 sl=64 p50 23,252 us 8,990 us 11,864 us +158.7% +96.0%
bs=10 sw=10 sl=64 p95 31,412 us 12,633 us 14,381 us +148.7% +118.4%
bs=10 sw=10 sl=64 p99 31,412 us 13,861 us 18,035 us +126.6% +74.2%
bs=100 sw=10 sl=64 throughput 946 tuples/sec 1,420 tuples/sec 1,083 tuples/sec -33.4% -12.7%
bs=100 sw=10 sl=64 MB/s 0.577 MB/s 0.867 MB/s 0.661 MB/s -33.4% -12.7%
bs=100 sw=10 sl=64 p50 105,361 us 69,686 us 93,077 us +51.2% +13.2%
bs=100 sw=10 sl=64 p95 130,178 us 75,137 us 99,553 us +73.3% +30.8%
bs=100 sw=10 sl=64 p99 130,178 us 89,601 us 108,604 us +45.3% +19.9%
bs=1000 sw=10 sl=64 throughput 1,103 tuples/sec 1,455 tuples/sec 1,119 tuples/sec -24.2% -1.4%
bs=1000 sw=10 sl=64 MB/s 0.673 MB/s 0.888 MB/s 0.683 MB/s -24.2% -1.4%
bs=1000 sw=10 sl=64 p50 907,832 us 685,976 us 909,247 us +32.3% -0.2%
bs=1000 sw=10 sl=64 p95 943,462 us 727,444 us 952,561 us +29.7% -1.0%
bs=1000 sw=10 sl=64 p99 943,462 us 767,738 us 985,186 us +22.9% -4.2%
Raw CSV
config_idx,batch_size,schema_width,string_len,num_batches,total_ms,total_tuples,total_bytes,tuples_per_sec,mb_per_sec,lat_p50_us,lat_p95_us,lat_p99_us
0,10,10,64,20,477.11,200,128000,419,0.256,23251.91,31411.91,31411.91
1,100,10,64,20,2115.17,2000,1280000,946,0.577,105361.45,130178.13,130178.13
2,1000,10,64,20,18133.80,20000,12800000,1103,0.673,907832.22,943461.62,943461.62

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

ddl-change Changes to the TexeraDB DDL engine

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants