Summary: Ref T11773. This is an initial first step toward a more complete solution, but should make the worst case much less bad: prior to this change, the worst case was "30 second exeuction timeout". After this patch, the worst case is "no results + explanatory message", which is strictly better.
Test Plan:
Made all feed stories fail policy checks, loaded home page.
- Before adding overheating: 9,600 queries / 20 seconds
- After adding overheating: 376 queries / 800ms
Reviewers: chad
Reviewed By: chad
Maniphest Tasks: T11773
Differential Revision: https://secure.phabricator.com/D16735
Summary:
Fixes T8911. This corrects several issues which could crop up if a calendar event query matched more results than the query limit:
- The desired order was not applied by the SearchEngine -- it applies the first builtin order instead. Provide a proper builtin order.
- When we generate ghosts, we can't do limiting in the database because we may select and then immediately discard a large number of parent events which are outside of the query range.
- For now, just don't limit results to get the behavior correct.
- This may need to be refined eventually to improve performance.
- When trimming events, we could trim parents and fail to generate ghosts from them. Separate parent events out first.
- Try to simplify some logic.
Test Plan: An "Upcoming" dashboard panel with limit 10 and the main Calendar "Upcoming Events" UI now show the same results.
Reviewers: chad
Reviewed By: chad
Maniphest Tasks: T8911
Differential Revision: https://secure.phabricator.com/D16289
Summary:
Ref T10537. These are objects which are bound to some external object, like a Maniphest task which is a representation of a GitHub issue.
This doesn't do much yet and may change, but my thinking is:
- I'm putting these on-object instead of on edges because I think we want to actively change the UI for them (e.g., clearly call out that the object is bridged) but don't want every page to need to do extra queries in the common case where zero bridged objects exist anywhere in the system.
- I'm making these one-to-one, more or less: an issue can't be bridged to a bunch of tasks, nor can a bunch of tasks be bridged to a single issue. Pretty sure this makes sense? I can't come up with any reasonable, realistic cases where you want a single GitHub issue to publish to multiple different tasks in Maniphest.
- Technically, one type of each bridgable object could be bridged, but I expect this to never actually occur. Hopefully.
Test Plan: Ran storage upgrade, loaded some pages.
Reviewers: chad
Reviewed By: chad
Subscribers: Luke081515.2
Maniphest Tasks: T10537
Differential Revision: https://secure.phabricator.com/D15502
Summary:
Ref T10188. If you issue certain queries which use reverse ordering (like "All tasks, oldest update to newest update") and then try to page forward, we build the paging clause without reversing the column order correctly.
For example, the ordering of "oldest update to newest update" is "dateModified ASC, id ASC", so the second page should include an "id > X" query. Currently, this builds as "id < X" incorrectly instead.
The cause of this is just a failure to re-reverse a reversing flag when constructing the paging clause.
Test Plan:
- Queried tasks by update, oldest to newest, with no grouping, etc.
- Paged to second page.
- After change, got a valid second page with a good query in the Services tab.
- Made some other normal queries.
Reviewers: chad
Reviewed By: chad
Maniphest Tasks: T10188
Differential Revision: https://secure.phabricator.com/D15076
Summary:
Fixes T10135. When the viewer is a member of no projects, specify the constraint type as a new "EMPTY" type.
When a query has an "EMPTY" constraint, fail fast with no results.
Test Plan:
- Viewed a viewerprojects() query result set as a user in no projects.
- Before patch: got a lot of hits. After patch: no hits.
- Viewed a normal result set, no changes.
Reviewers: chad
Reviewed By: chad
Maniphest Tasks: T10135
Differential Revision: https://secure.phabricator.com/D15003
Summary:
Ref T10010. I think this is the desired/expected default behavior (e.g., searching for "Maniphest" should find tasks in any subproject or sprint of that project).
I'll probably add an "exact(...)" function later to mean "only the Maniphest superproject, exactly, not any of its children".
Test Plan:
- Added and executed unit tests.
- Ran various queries from the web UI.
- Got sensible-seeming results.
Reviewers: chad
Reviewed By: chad
Maniphest Tasks: T10010
Differential Revision: https://secure.phabricator.com/D14910
Summary:
Ref T9979. This uses ngrams (specifically, trigrams) to build a reasonably efficient index for substring matching. Specifically, for a package like "Example", with ID 123, we store rows like this:
```
< ex, 123>
<exa, 123>
<xam, 123>
<amp, 123>
<mpl, 123>
<ple, 123>
<le , 123>
```
When the user searches for `exam`, we join this table for packages with tokens `exa` and `xam`. MySQL can do this a lot more efficiently than it can process a `LIKE "%exam%"` query against a huge table.
When the user searches for a one-letter or two-letter string, we only search the beginnings of words. This is probably what they want, the only thing we can do quickly, and a reasonable/expected behavior for typeaheads.
Test Plan:
- Ran storage upgrades and search indexer.
- Searched for stuff with "name contains".
- Used typehaead and got sensible results.
- Searched for `aabbccddeeffgghhiijjkkllmmnnooppqqrrssttuuvvwwxxyyzz` and saw only 16 joins.
Reviewers: chad
Reviewed By: chad
Maniphest Tasks: T9979
Differential Revision: https://secure.phabricator.com/D14846
Summary:
Fixes T6864. This creates a sort of busy menu but I think that's proably fine -- users are opting into activating these fields for search anyway.
In the future, we could refine this as, e.g.:
- don't show these options in the dropdown;
- do show them on some new "http prefilling" sort of page;
- then you access them as an advanced user with `?order=secret-magic`.
But I'm not going to bother for now.
Test Plan: Ordered by an int field, then reversed the order.
Reviewers: chad
Reviewed By: chad
Maniphest Tasks: T6864
Differential Revision: https://secure.phabricator.com/D14800
Summary:
Ref T9964. Adds a new-style "owners.search" endpoint, and an extension for customfields.
Puts enough indirection in place to give us nice, consistent "custom.key" user-facing keys instead of "std:custom:owners:na0shf9a8dfdsafl" junk.
Test Plan:
- Searched Owners via API.
- Searched by ID.
- Ordered by custom fields.
- Reviewed API docs.
- Used normal search with ordering.
- Viewed custom field values in search results.
Reviewers: chad
Reviewed By: chad
Maniphest Tasks: T9964
Differential Revision: https://secure.phabricator.com/D14758
Summary:
Fixes T8743. Fixes T8746. When running queries with the omnipotent viewer and no explicit space constraints, don't add implicit space constraints.
This prevents us from fataling when running older pre-space migrations and trying to load space-aware objects.
Test Plan: Manually ran migrations with `--trace`, verified no `WHERE spacePHID = ...`.
Reviewers: btrahan, chad
Reviewed By: chad
Subscribers: eadler, epriestley
Maniphest Tasks: T8743, T8746
Differential Revision: https://secure.phabricator.com/D13542
Summary:
Ref T8731. I think the issue is that some `ProjectQuery` (without needImages()) populates the query workspace, then the "real" one hits the workspace.
Instead, only populate the workspace from ObjectQuery, so we know that objects in the workspace always have whatever ObjectQuery attaches to them.
Test Plan: Verified this didn't destroy the cache hitrate, but I can't repro the original issue locally per se.
Reviewers: btrahan
Reviewed By: btrahan
Subscribers: chad, epriestley
Maniphest Tasks: T8731
Differential Revision: https://secure.phabricator.com/D13516
Summary:
Ref T8631. The query plan for feed stories is really bad right now, because we miss caches we should be hitting:
- The workspace cache is stored at each query, so adjacent queries can't benefit from the cache (only subqueries). Feed has primarily sibling queries.
- There is no technical reason to do this. Store the workspace cache on the root query, so sibling queries can hit it.
- In `ObjectQuery`, we check the workspace once, then load all the PHIDs. When the PHIDs are a mixture of transactions and objects, we always miss the workspace and load the objects twice.
- Instead, check the workspace after loading each type of object.
- `HandleQuery` does not set itself as the parent query for `ObjectQuery`, so handles never hit the workspace cache.
- Pass it, so they can hit the workspace cache.
- Feed's weird `PhabricatorFeedStory::loadAllFromRows()` method does not specify a parent query on its object/handle queries.
- Just declare the object query to be the "root" query until this eventually gets cleaned up.
Test Plan: Saw queries for each object drop from 4-6x to 1x in `/feed/`.
Reviewers: btrahan
Reviewed By: btrahan
Subscribers: epriestley
Maniphest Tasks: T8631
Differential Revision: https://secure.phabricator.com/D13479
Summary:
Ref T8575. Because orderable columns need to build custom fields, they are relatively expensive to build. Use the request cache.
(The request cache is technically more correct than `static`, because configuration may change between requests and we may eventually reuse interpreters.)
Test Plan: Saw home page time drop 39% (from 462ms to 283ms).
Reviewers: btrahan, avivey
Reviewed By: avivey
Subscribers: avivey, epriestley
Maniphest Tasks: T8575
Differential Revision: https://secure.phabricator.com/D13322
Summary: Ref T8575. We check if users can see applications frequently, and caching on the Query isn't especially effective. Use the new Request cache instead.
Test Plan:
- Saw `/feed/` drop 7% (from ~830ms to ~770ms) on profiles.
Reviewers: btrahan, avivey
Reviewed By: avivey
Subscribers: avivey, epriestley
Maniphest Tasks: T8575
Differential Revision: https://secure.phabricator.com/D13321
Summary: All classes should extend from some other class. See D13275 for some explanation.
Test Plan: `arc unit`
Reviewers: epriestley, #blessed_reviewers
Reviewed By: epriestley, #blessed_reviewers
Subscribers: epriestley, Korvin
Differential Revision: https://secure.phabricator.com/D13283
Summary:
Ref T4345. This error is per object-type in the query implementations, not a mail/permissions issue.
Without `didRejectResult()`, we can't distinguish between "restricted" and "unknown" for objects filtered by `willFilterPage()`.
- Call `didRejectResult()` on commits.
- Make `didRejectResult()` handle both existing policy exceptions and filtering.
- Recover from partial objects (like commits) which are missing attached data required to figure out policies.
Test Plan: Saw "Restricted Diffusion Commit" instead of "Unknown Object (Diffusion Commit)" when viewing nonvisible commit handle in Maniphest.
Reviewers: btrahan, joshuaspence
Reviewed By: joshuaspence
Subscribers: epriestley
Maniphest Tasks: T4345
Differential Revision: https://secure.phabricator.com/D13289
Summary:
Ref T8377. This adds a standard disable/enable feature to Spaces, with a couple of twists:
- You can't create new stuff in an archived space, and you can't move stuff into an archived space.
- We don't show results from an archived space by default in ApplicationSearch queries. You can still find these objects if you explicitly search for "Spaces: <the archived space>".
So this is a "put it in a box in the attic" sort of operation, but that seems fairly nice/reasonable.
Test Plan:
- Archived and activated spaces.
- Used ApplicationSearch, which omitted archived objects by default but allowed searches for them, specifically, to succeed.
- Tried to create objects into an archived space (this is not allowed).
- Edited objects in an archived space (this is OK).
Reviewers: btrahan
Reviewed By: btrahan
Subscribers: epriestley
Maniphest Tasks: T8377
Differential Revision: https://secure.phabricator.com/D13238
Summary: Ref T8441. Ref T7715. Ref T7909. Clean up all the ordering and grouping hacks in Maniphest so we can drive it through normal infrastructure, move it to SearchField, introduce Spaces, and eventually modernize the Conduit API.
Test Plan:
- Executed all grouping/ordering queries, including custom queries.
- Forced execution with old aliases; got modern results.
Reviewers: btrahan
Reviewed By: btrahan
Subscribers: epriestley
Maniphest Tasks: T7909, T7715, T8441
Differential Revision: https://secure.phabricator.com/D13197
Summary:
Ref T8441. Ref T7715. Automatically generate a modern "Order" control in ApplicationSearch for engines which fully support SearchField.
Notably, this allows the standard "Order" control to automatically support custom field orders. We do this in Maniphest today, but in an ad-hoc way.
Test Plan: Performed order-by queries in Almanac (Services), Pholio, Files, People, Projects, and Paste.
Reviewers: btrahan
Reviewed By: btrahan
Subscribers: epriestley
Maniphest Tasks: T7715, T8441
Differential Revision: https://secure.phabricator.com/D13193
Summary:
Ref T8441. Ref T7715. For modern Query classes, automatically make subscriber queries and SearchField integrations work.
In particular, we can just drive this query with EdgeLogic and don't need to do anything specific on these Query classes beyond making sure they're implemented in a way that picks up all of the EdgeLogic clauses.
Test Plan:
- Searched for subscribers in Pholio, Files, Paste, and Projects.
- Searched for all other fields in Projects to check that Query changes are OK.
Reviewers: btrahan
Reviewed By: btrahan
Subscribers: epriestley
Maniphest Tasks: T7715, T8441
Differential Revision: https://secure.phabricator.com/D13191
Summary:
Ref T8424. No UI or interesting behavior yet, but integrates Spaces checks:
- `PolicyFilter` now checks Spaces.
- `PolicyAwareQuery` now automatically adds Spaces constraints.
There's one interesting design decision here: **spaces are stronger than automatic capabilities**. That means that you can't see a task in a space you don't have permission to access, //even if you are the owner//.
I //think// this is desirable. Particularly, we need to do this in order to exclude objects at the query level, which potentially makes policy filtering for spaces hugely more efficient. I also like Spaces being very strong, conceptually.
It's possible that we might want to change this; this would reduce our access to optimizations but might be a little friendlier or make more sense to users later on.
For now, at least, I'm pursuing the more aggressive line. If we stick with this, we probably need to make some additional UI affordances (e.g., show when an owner can't see a task).
This also means that you get a hard 404 instead of a policy exception when you try to access something in a space you can't see. I'd slightly prefer to show you a policy exception instead, but think this is generally a reasonable tradeoff to get the high-performance filtering at the Query layer.
Test Plan:
- Added and executed unit tests.
- Put objects in spaces and viewed them with multiple users.
- Made the default space visible/invisible, viewed objects.
- Checked the services panel and saw `spacePHID` constraints.
- Verified that this adds only one query to each page.
Reviewers: btrahan, chad
Reviewed By: btrahan
Subscribers: chad, epriestley
Maniphest Tasks: T8424
Differential Revision: https://secure.phabricator.com/D13156
Summary: Ref T8099, Moves AphrontPagerView to PHUIPagerView, converts to standard PHUIButtons and adds some additional features for icon placement on buttons.
Test Plan: Tested Advanced Search and Searching files in Diffusion. Works as expected.
Reviewers: btrahan, epriestley
Reviewed By: epriestley
Subscribers: Korvin, epriestley
Maniphest Tasks: T8342, T8099
Differential Revision: https://secure.phabricator.com/D13092
Summary:
Fixes T6726. Currently, a file may be attached to itself (or to other files, ultimately forming a loop). In this case, we currently run around the loop forever trying to load all the files.
Instead, decline to load objects if we're inside a query which is already loading them. This produces the right policy result //and// completes in finite time.
Test Plan:
- Looped two files by writing `{F123}` and `{F124}` on the other files, respectively.
- Loaded `F123`.
- Saw long hang; used `debug.time-limit` to see huge stack trace instead.
- Wrote patch.
- `F123` now loads correctly.
Reviewers: btrahan
Reviewed By: btrahan
Subscribers: epriestley
Maniphest Tasks: T6726
Differential Revision: https://secure.phabricator.com/D12756
Summary: See IRC. This got dropped in the order refactoring.
Test Plan: Ordered Maniphest search results by a custom field.
Reviewers: btrahan
Reviewed By: btrahan
Subscribers: epriestley
Differential Revision: https://secure.phabricator.com/D12614
Summary:
Ref T4100. Share all edge logic code across applications.
- Internalizes the "check that the viewer can see projects" check into edge logic.
- Adds some convenience functions. Some of these aren't really all that convenient, but it's rare that we actually apply project constraints to queries in the applications -- and most of these callsites will go away in the long term -- so I didn't go too crazy with providing a simpler `withProjectPHIDs()` universal API or anything.
Test Plan:
- Grepped for all affected symbols.
- Tried to violate policies.
- Used workboards.
- Used normal Maniphest queries.
- Used `maniphest.query`.
- Verified the special grouping behavior works as expected.
Reviewers: btrahan
Reviewed By: btrahan
Subscribers: epriestley
Maniphest Tasks: T4100
Differential Revision: https://secure.phabricator.com/D12526
Summary:
Ref T4100. Ref T5595. This implements these fields in one mega-field:
- Projects
- Not in projects
- In any project
- Include results in no projects
- In users' projects
Hopefully, this is a step in the right direction.
Test Plan: {F375555}
Reviewers: btrahan
Reviewed By: btrahan
Subscribers: joshuaspence, chad, epriestley
Maniphest Tasks: T4100, T5595
Differential Revision: https://secure.phabricator.com/D12463
Summary:
Ref T4100. Ref T5595. This allows PolicyAwareQuery to write all the logic for AND, OR, NOT, and NULL (i.e., "not in any projects") queries against any edge type.
It accepts an edge type and a list of constraints (which are basically just operator-value pairs, like `<NOT, PHID-X-Y>`, meaning the results must not have an edge connecting them to `PHID-X-Y`).
This doesn't actually do anything yet; see future diffs.
Test Plan: `arc unit --everything`
Reviewers: btrahan
Reviewed By: btrahan
Subscribers: epriestley
Maniphest Tasks: T4100, T5595
Differential Revision: https://secure.phabricator.com/D12455
Summary:
Ref T4100. Ref T5595. These functions are trivial for now, but move us toward being able to define more default query behavior by default.
Future changes will give these methods meaningful, nontrivial behaviors.
Test Plan: `arc unit --everything`
Reviewers: btrahan
Reviewed By: btrahan
Subscribers: epriestley
Maniphest Tasks: T5595, T4100
Differential Revision: https://secure.phabricator.com/D12454
Summary:
Ref T4100. Ref T5595.
To support a unified "Projects:" query across all applications, a future diff is going to add a set of "Edge Logic" capabilities to `PolicyAwareQuery` which write the required SELECT, JOIN, WHERE, HAVING and GROUP clauses for you.
With the addition of "Edge Logic", we'll have three systems which may need to build components of query claues: ordering/paging, customfields/applicationsearch, and edge logic.
For most clauses, queries don't currently call into the parent explicitly to get default components. I want to move more query construction logic up the class tree so it can be shared.
For most methods, this isn't a problem, but many subclasses define a `buildWhereClause()`. Make all such definitions protected and consistent.
This causes no behavioral changes.
Test Plan: Ran `arc unit --everything`, which does a pretty through job of verifying this statically.
Reviewers: btrahan
Reviewed By: btrahan
Subscribers: yelirekim, hach-que, epriestley
Maniphest Tasks: T4100, T5595
Differential Revision: https://secure.phabricator.com/D12453
Summary: Ref T5750. Update the Almanac service query to be browsable.
Test Plan:
- Browsed and reordered Diffusion.
- Browsed and reordered services in Almanac.
{F373735}
Reviewers: btrahan
Reviewed By: btrahan
Subscribers: epriestley
Maniphest Tasks: T5750
Differential Revision: https://secure.phabricator.com/D12433
Summary:
Ref T7803. Prior to this change sequence, Query classes conflated paging values (the actual thing that goes in a "x > 3" clause) with cursor values (arbitrary identifiers which track where the user is in a result list).
Although the two can sometimes be the same, the vast majority of implementations are simpler and better when object IDs are used as cursors and paging values are derived from them.
The new stuff handles this in a consistent way, so we're free to separate getPagingValue() from paging. The new method is essentially getResultCursor().
This also implements getPageCursors(), which allows queries to return directional cursors. The inability to do this was a practical limitation blocking the implementation of T7803.
Test Plan:
- Browsed a bunch of results and paged through queries.
- Grepped for removed methods.
Reviewers: btrahan
Reviewed By: btrahan
Subscribers: epriestley
Maniphest Tasks: T7803
Differential Revision: https://secure.phabricator.com/D12383
Summary:
Ref T7803. Currently, available high-level orders are spread across Query and SearchEngine classes and implemented separately for each application.
Lift the concept of "builtin" (high-level, user-facing, named) orders (similar to "builtin" queries in ApplicationSearch) into the root Query class, and let it drive the SearchEngine implementation. This allows you to define a new order in one place and have it automatically work across the entire stack.
This will also let Conduit expose this information in a straightforward way.
Test Plan:
- Used ApplicationSearch in Diffusion.
- Used all result orderings.
Reviewers: btrahan
Reviewed By: btrahan
Subscribers: epriestley
Maniphest Tasks: T7803
Differential Revision: https://secure.phabricator.com/D12379
Summary: Ref T7803. Remove these in favor of more generalized paging and ordering.
Test Plan: Sorted and paged results in various applications.
Reviewers: btrahan
Reviewed By: btrahan
Subscribers: epriestley
Maniphest Tasks: T7803
Differential Revision: https://secure.phabricator.com/D12378
Summary:
Ref T7803. The ApplicationSearch integration is still a little rough here, but it seems to have the correct behavior.
The rest of this is now at least relatively sane, cohesive, and properly behaved.
Test Plan:
- Used all grouping and ordering queries in Maniphest. Pagingated results.
- Used custom field ordering in Maniphest. Paginated results.
- Paginated through the `null` section of "Assigned" and "Projects" group-by queries. Pagingation now works correctly (it does not work at HEAD).
- Ran unit tests covering priority changes.
Reviewers: btrahan
Reviewed By: btrahan
Subscribers: epriestley
Maniphest Tasks: T7803
Differential Revision: https://secure.phabricator.com/D12372
Summary:
Ref T7803. Removes some getReversePaging().
This also fixes `null` column handling, by adding an explicit `'null'` key with possible values "head" (put NULL before other values) or "tail" (put NULL after other values).
Maniphest has some glitchiness in paging through NULLs right now, but I believe it's all pre-existing and will be resolved when it fully converts. Diffusion is fully converted and pages through NULL correctly.
Test Plan:
- Failed to identify any reason for ChangesetQuery to reverse paging.
- Paged thorugh Diffusion.
- Paged through Maniphest.
- Maniphest has some issues when paging inside a NULL section, but these issues are preexisting and will be resolved later in this change sequence.
Reviewers: btrahan
Reviewed By: btrahan
Subscribers: epriestley
Maniphest Tasks: T7803
Differential Revision: https://secure.phabricator.com/D12371
Summary:
Ref T7803. Some Query subclasses implement getPagingColumn() in a trivial way, usually to provide a table alias.
Formalize the concept of a primary table alias, and remove obsoleted getPagingColumn() implementations.
Test Plan: Issued affected queries.
Reviewers: btrahan
Reviewed By: btrahan
Subscribers: epriestley
Maniphest Tasks: T7803
Differential Revision: https://secure.phabricator.com/D12356
Summary:
Ref T7803. Ordering and paging are inherently intertwined, but they often aren't driven by the same data right now.
Start driving them through the same data:
- `getOrderableColumns()` defines orderable and pageable columns.
- `getPagingValueMap()` reads values from a cursor.
This is generally sufficient to implement both paging and ordering.
Also, add some more sanity checks to try to curtail the number of ambiguous/invalid orderings applications produce, since these cause subtle/messy bugs.
Test Plan:
- Paged through pastes and a few other object types.
- Intentionally changed defaults to be invalid and hit some of the errors.
Reviewers: btrahan
Reviewed By: btrahan
Subscribers: epriestley
Maniphest Tasks: T7803
Differential Revision: https://secure.phabricator.com/D12355
Summary: Ref T7803. Reduce the amount of code we're trusting to build SQL queries.
Test Plan:
- Paged through results in Maniphest, Differential and Diffusion.
- Some of the NULLable groups in Maniphest are a bit funky but this was preexisting.
Reviewers: btrahan
Reviewed By: btrahan
Subscribers: epriestley
Maniphest Tasks: T7803
Differential Revision: https://secure.phabricator.com/D12353
