[data.search] Clean up arguments to esaggs. #84973

lukeelmers · 2020-12-04T00:03:28Z

Closes #61768
Closes #65067
Part of #46902

Summary

This updates the arguments of esaggs a part of the project to clean up semantics across all expression functions by removing blobs of stringified json. The following changes have been made here:

Updates the index arg in esaggs to require an argument of type index_pattern which is only returnable by indexPatternLoad
Updates the aggConfigs arg to be an aggs multiarg, which accepts type agg_type which is only returnable by each of the various agg type functions
Updates downstream usage of esaggs in both Lens & Visualize to ensure the function is called correctly
- In Lens, updated the toEsAggsConfig datasource method to be toEsAggsFn and return the function AST. This is basically what @wylieconlon did with his exploration in [Lens] Use agg functions instead of string #77565
- In Visualize, updated build_pipeline as well as each of the vis types that already had their own toExpressionAst defined
Cleans up a few bugs that were uncovered along the way in some of the agg type functions & in aggConfig.toExpressionAst()
Cleans up a bug I found in visualize which appears to be a regression introduced in Vislib visualization renderer #80744 (not yet released).
- The problem was with the serial diff agg, which is currently broken on master.
Updates the esaggs implementation & tests
Adds unit tests for esaggs & requestHandler.
Adds functional tests to the interpreter functional test suite for server-run expressions.

Testing

This PR introduces no functional changes, however it makes significant changes to the underlying infrastructure for generating aggregation requests that are run for Visualize & Lens. The best things to test are various configurations of agg types in different visualizations to make sure things are behaving as expected. Ultimately no user-facing behavior should change as a result of this PR.

Plugin API Changes

The esaggs expression function, which is the default method of requesting aggregations for visualizations, has had some changes to the arguments it accepts.

// old
esaggs index="logstash-*" aggConfigs="[{ id: 1, enabled: true, type: "count", schema: "metric" }]"

// new
esaggs
  // use indexPatternLoad and pass your ID instead of passing it as a string
  index={indexPatternLoad id="logstash-*"}
  // use aggType functions for each aggregation you need. the aggs argument
  // can be passed multiple times. if you are using AggConfigs you can automatically
  // generate the expression AST for these arguments with `aggConfig.toExpressionAst()`
  aggs={aggCount id=1 enabled=true schema="metric"}

flash1293 · 2020-12-04T14:48:52Z

x-pack/plugins/lens/public/indexpattern_datasource/operations/definitions/cardinality.tsx

-    schema: 'metric',
-    params: {
+  toEsAggsFn: (column, columnId) => {
+    return buildExpressionFunction<AggFunctionsMapping['aggCardinality']>('aggCardinality', {


This approach looks great to me, seems like that's just as maintainable as the old version 👍

seems like that's just as maintainable as the old version

Yeah after our discussion I realized that, once converting to use the function builder, you gain a lot more type safety than you had previously, so overall I think that will help prevent any future issues 🙂

lukeelmers · 2020-12-07T04:28:07Z

src/plugins/data/common/search/aggs/agg_config.test.ts

+    it('does not stringify arrays which are not objects', () => {
+      const ac = new AggConfigs(indexPattern, [], { typesRegistry });
+      const configStates = {
+        type: 'percentiles',
+        params: {
+          field: 'bytes',
+          percents: [1, 25, 50, 75, 99],
+        },
+      };
+      const aggConfig = ac.createAggConfig(configStates);
+      const percents = aggConfig.toExpressionAst()?.chain[0].arguments.percents;
+      expect(percents).toEqual([1, 25, 50, 75, 99]);
+    });


Uncovered a bug in toExpressionAst where the percents arg was incorrectly stringified when we should have preserved it, so I updated the logic to handle this case where a param is already an array, but not an object array.

lukeelmers · 2020-12-07T04:28:43Z

src/plugins/data/common/search/aggs/agg_config.ts

   */
-  toExpressionAst(): ExpressionAstFunction | undefined {
+  toExpressionAst(): ExpressionAstExpression | undefined {


Discovered that returning an ExpressionAstFunction was not very useful since using this as a subexpression requires it to be ExpressionAstExpression, so I updated accordingly.

lukeelmers · 2020-12-07T04:29:24Z

src/plugins/data/common/search/aggs/agg_types.ts

-import { getMovingAvgMetricAgg } from './metrics/moving_avg';
-import { getSerialDiffMetricAgg } from './metrics/serial_diff';
+import * as buckets from './buckets';
+import * as metrics from './metrics';


No changes here, this file was just cleanup.

lukeelmers · 2020-12-07T04:30:43Z

src/plugins/data/common/search/aggs/buckets/histogram_fn.ts

@@ -72,7 +72,7 @@ export const aggHistogram = (): FunctionDefinition => ({
      }),
    },
    interval: {
-      types: ['string'],
+      types: ['number', 'string'],


We had incorrectly typed this as string only (since you can do auto), but in most cases the interval is actually a number when explicitly provided. Keeping this as a string caused some functional test failures where an incorrect interval got applied when the histogram expression function cast the provided number to a string instead of keeping its type.

lukeelmers · 2020-12-07T04:32:17Z

src/plugins/data/common/search/aggs/buckets/lib/histogram_calculate_interval.ts

+  let calculatedInterval = isAuto
+    ? 0
+    : typeof interval !== 'number'
+    ? parseFloat(interval)
+    : interval;


Had to tweak logic here to make TS happy once updating interval to be string or number. Now we need to check for the case where it's already a number.

Technically this doesn't change anything at runtime (since passing a number to parseFloat will still return the same number), but TS expects parseFloat to only accept a number.

lukeelmers · 2020-12-07T04:36:58Z

src/plugins/data/public/search/expressions/esaggs.ts

-      const aggConfigsState = JSON.parse(args.aggConfigs);
-      const indexPattern = await indexPatterns.get(args.index);
-      const aggConfigs = aggs.createAggConfigs(indexPattern, aggConfigsState);
+      const indexPattern = await indexPatterns.create(args.index.value, true);


I used indexPatterns.create here, because manually creating new IndexPattern() still requires you to provide the field formats service, plus some UI settings... which means adding more dependencies to esaggs (currently ui settings is not required).

However we could still refactor to use new IndexPattern() directly if desired; this was just a simpler change.

lukeelmers · 2020-12-07T04:44:39Z

src/plugins/vis_type_vislib/public/to_ast.ts

-  const configStr = JSON.stringify(visConfig).replace(/\\/g, `\\\\`).replace(/'/g, `\\'`);
  const visTypeXy = buildExpressionFunction<VisTypeVislibExpressionFunctionDefinition>(
    vislibVisName,
    {
      type: vis.type.name,
-      visConfig: configStr,


There was a regression here which AFAICT was introduced by #80744.

It surfaced as a bug where creating a vislib vis with a serial diff agg threw an error... Because the custom label that got created had incorrectly double-escaped backslashes, e.g.: "Blah blah serial diff of \\"bytes\\""

This would cause JSON.parse to throw because the double-escaped quotes were accidentally kept in place and interpreted as part of the json, unexpectly ending the line early.

I know this escaping was copied over from build_pipeline and TBH I'm not 100% sure why it isn't needed anymore, but thought I'd provide some background. I only found this bug because I was testing visualizations using every agg type to validate this PR, and ran into trouble with serial diff.

Also worth pointing out that the to_ast for pie charts still has the escaping in place, because i couldn't find a situation where it was broken. Might be worth testing that further to verify.

I tested both and I think that they both work as they should. I tested all the aggs on all vislib visualizations. Can't find any regressions.

lukeelmers · 2020-12-07T04:47:08Z

x-pack/plugins/lens/public/indexpattern_datasource/operations/definitions/index.ts

@@ -4,7 +4,6 @@
 * you may not use this file except in compliance with the Elastic License.
 */

-import { ExpressionFunctionAST } from '@kbn/interpreter/common';


As an aside -- At some point y'all should consider removing all of the old @kbn/interpreter imports from Lens since you can access the same types (with slightly different names) from expressions.

I just did so in the files I touched here

lukeelmers · 2020-12-07T04:49:52Z

x-pack/plugins/lens/public/indexpattern_datasource/operations/definitions/metrics.tsx

        field: column.sourceField,
-        missing: 0,


I don't believe missing is a valid param for any of these agg types so I removed it; please LMK if you know otherwise.

lukeelmers · 2020-12-07T04:56:29Z

x-pack/plugins/lens/public/indexpattern_datasource/operations/definitions/ranges/ranges.tsx

+      maxBars: params.maxBars === AUTO_BARS ? undefined : params.maxBars,
+      interval: 'auto',
+      has_extended_bounds: false,
+      min_doc_count: true,


Important: Please confirm that the update I made to min_doc_count doesn't have any unexpected effects on Lens, because I think it might actually be changing the behavior...

According to the histogram agg type, min_doc_count should be a boolean (not a number). It writes out the param as follows:

if (aggConfig.params.min_doc_count) { output.params.min_doc_count = 0; } else { output.params.min_doc_count = 1; }

So as you can see, setting it to true will return a value of 0, which I why I changed it to true here as it was previously set to 0.

However, since the check is just confirming whether min_doc_count is truthy, 0 might instead cause it to actually be set to 1 (or rather, the expressions interpreter could be casting it to false). So it's possible that lens was inadvertently relying on this behavior, which is why I'd appreciate a second set of eyes on this one.

Good point, yes we are relying on this behavior, could you flip the boolean?

Otherwise histogram is hard to use on a table:

elasticmachine · 2020-12-07T04:59:15Z

Pinging @elastic/kibana-app-services (Team:AppServices)

lukeelmers · 2020-12-09T04:03:29Z

src/plugins/expressions/common/execution/execution.ts

+const getNewAbortController = (): AbortController => {
+  try {
+    return new AbortController();
+  } catch (error) {
+    // eslint-disable-next-line @typescript-eslint/no-var-requires
+    const polyfill = require('abortcontroller-polyfill/dist/cjs-ponyfill');
+    return new polyfill.AbortController();
+  }


This was a bug uncovered by functional tests which we hadn't found yet: Since AbortController isn't supported till Node 15 (and we are currently on v14), we need to manually apply a polyfill on the server.

Since this code is in common, I'm hacking around the polyfill with a try/catch.

lukeelmers · 2020-12-09T04:04:47Z

src/plugins/data/server/search/search_service.ts

@@ -254,8 +254,7 @@ export class SearchService implements Plugin<ISearchSetup, ISearchStart> {
          const searchSourceDependencies: SearchSourceDependencies = {
            getConfig: <T = any>(key: string): T => uiSettingsCache[key],
            search: asScoped(request).search,
-            // onResponse isn't used on the server, so we just return the original value
-            onResponse: (req, res) => res,
+            onResponse: (req, res) => shimHitsTotal(res),


I discovered that shimHitsTotal is applied in the low-level search routes, but not in the actual service that's available on the plugin contract, so we need to apply it here for search source.

(This is only necessary on the server, as the client is going through the http routes)

cc @lizozom @lukasolson Whenever shimHitsTotal is updated to be applied at the search strategy level instead of the route level, this change can be reverted.

lukeelmers · 2020-12-09T04:06:32Z

test/interpreter_functional/test_suites/run_pipeline/esaggs.ts

  let expectExpression: ExpectExpression;
+
+  const expectClientToMatchServer = async (title: string, expression: string) => {


I added a function to compare results from client & server. Longer term we might consider updating helpers.ts to have expectExpressionServer or similar, which would make it even easier to run expressions on the server.

lukeelmers · 2020-12-09T04:07:50Z

test/interpreter_functional/plugins/kbn_tp_run_pipeline/server/plugin.ts

+
+    router.post(
+      {
+        path: '/api/interpreter_functional/run_expression',


All this route does is take an expression & optional input, call expressions.run from the server contract, and return the result.

This way we can gradually build out a suite that tests the same expressions on both client & server.

ppisljar

code LGTM

flash1293

Found some small things, but overall it looks good

flash1293 · 2020-12-09T14:55:39Z

x-pack/plugins/lens/public/indexpattern_datasource/to_expression.ts

-            index: [indexPattern.id],
-            metricsAtAllLevels: [false],
-            partialRows: [false],
-            includeFormatHints: [true],


Just to make sure - did the flag includeFormatHints become unnecessary?

Yes, includeFormatHints isn't even returned in the esaggs response on master -- the argument itself just never got removed.

x-pack/plugins/lens/public/indexpattern_datasource/operations/definitions/ranges/ranges.tsx

x-pack/plugins/lens/server/migrations.test.ts

lukeelmers · 2020-12-09T18:15:49Z

@elasticmachine merge upstream

kibanamachine · 2020-12-09T19:56:22Z

💚 Build Succeeded

continuous-integration/kibana-ci/pull-request
Commit: 35a6923

Metrics [docs]

Module Count

Fewer modules leads to a faster build time

id	before	after	diff
`expressions`	103	104	+1

Async chunks

Total size of all lazy-loaded chunks that will be downloaded as the user navigates the app

id	before	after	diff
`data`	244.3KB	244.3KB	-2.0B
`lens`	1011.8KB	1012.5KB	+699.0B
total			+697.0B

Distributable file count

id	before	after	diff
`default`	46984	47744	+760

Page load bundle

Size of the bundles that are downloaded on every page load. Target size is below 100kb

id	before	after	diff
`data`	985.1KB	1002.4KB	+17.3KB
`expressions`	182.6KB	189.9KB	+7.3KB
`lens`	53.2KB	53.2KB	+7.0B
`visTypeMetric`	25.9KB	26.0KB	+163.0B
`visTypeTable`	18.9KB	19.1KB	+163.0B
`visTypeTagcloud`	18.8KB	19.0KB	+163.0B
`visTypeVislib`	66.0KB	66.1KB	+91.0B
`visualizations`	169.0KB	169.2KB	+162.0B
total			+25.3KB

History

💔 Build #93166 failed d4037ed
💚 Build #92983 succeeded 18f62e0
💔 Build #92954 failed bf57656077f73e0f9dafff03b286e050e05d33c2
💚 Build #92924 succeeded de890200695778b413e2ff3ee3eeddf30c5fa751
💚 Build #92565 succeeded 7d2c3db939cd53c9f27dcf1af8753a5ba4736535

To update your PR or re-run it, just comment with:
@elasticmachine merge upstream

flash1293

LGTM, thanks for making this change and giving us some type safety around agg building :) There are some JSON-serialized parts in there I guess you will revisit them later? Let me know if the Lens team can help here.

I thought a bit more about the histogram min_doc_count thing and I think we have to do some fixes here, but for this PR it makes sense to keep the existing behavior even if it's not perfect.

flash1293 · 2020-12-10T08:53:49Z

I focused on Lens, would appreciate a look at the Visualize bits from @stratoula

stratoula

Legacy visualizations code review LGTM, I tested all of them with various aggs and I can't find any regression. Thanx @lukeelmers for fixing the serial Diff bug 🙂

lukeelmers · 2020-12-10T14:38:45Z

There are some JSON-serialized parts in there I guess you will revisit them later? Let me know if the Lens team can help here.

@flash1293 Yep we have #67908 to track further work in this area. I'm sure the team would love any help you can provide 😄

Each agg param type now has its own toExpressionAst which is called by the top-level aggConfig.toExpressionAst, so the aim is to eventually add expression functions for each of these, and update their toExpressionAst methods accordingly.

lukeelmers added the WIP Work in progress label Dec 4, 2020

lukeelmers force-pushed the feat/esaggs-args branch from ad36054 to 6b2eb4c Compare December 4, 2020 05:43

flash1293 reviewed Dec 4, 2020

View reviewed changes

lukeelmers force-pushed the feat/esaggs-args branch 3 times, most recently from ec49568 to efb40b0 Compare December 7, 2020 04:26

lukeelmers commented Dec 7, 2020

View reviewed changes

lukeelmers self-assigned this Dec 7, 2020

lukeelmers requested review from ppisljar and wylieconlon December 7, 2020 04:58

lukeelmers added review and removed WIP Work in progress labels Dec 7, 2020

lukeelmers marked this pull request as ready for review December 7, 2020 04:59

lukeelmers requested a review from a team December 7, 2020 04:59

lukeelmers requested a review from a team as a code owner December 7, 2020 04:59

lukeelmers force-pushed the feat/esaggs-args branch from efb40b0 to 7d2c3db Compare December 7, 2020 23:09

botelastic bot added the Feature:Embedding Embedding content via iFrame label Dec 8, 2020

lukeelmers force-pushed the feat/esaggs-args branch 2 times, most recently from a668d50 to bf57656 Compare December 9, 2020 00:00

lukeelmers added 3 commits December 8, 2020 20:58

Add i18n for argument help text.

83e9d4b

Remove includeFormatHint arg.

f86623e

Update index arg to use indexPatternLoad subexpression.

2a879c2

lukeelmers added 8 commits December 8, 2020 20:58

Fix vislib regression from elastic#80744.

256d7b1

Update lens datasource to build aggs as subexpressions.

782696e

Update interpreter functional tests.

e739485

Add esaggs & request handler unit tests.

96d12d4

Update generated docs.

7a07b7f

Fix bug where we didn't shim total hits in search source on the server.

8e258ec

Fix bug where AbortController was not polyfilled on the server.

1e086ee

Add interpreter functional tests on the server.

18f62e0

lukeelmers force-pushed the feat/esaggs-args branch from bf57656 to 18f62e0 Compare December 9, 2020 03:58

lukeelmers requested a review from flash1293 December 9, 2020 03:59

lukeelmers commented Dec 9, 2020

View reviewed changes

ppisljar approved these changes Dec 9, 2020

View reviewed changes

flash1293 requested changes Dec 9, 2020

View reviewed changes

lukeelmers added 2 commits December 9, 2020 09:04

Revert lens migration test changes.

061748a

Ensure min_doc_count set to false in lens histograms.

d4037ed

lukeelmers requested a review from flash1293 December 9, 2020 16:40

Merge branch 'master' into feat/esaggs-args

35a6923

flash1293 approved these changes Dec 10, 2020

View reviewed changes

stratoula approved these changes Dec 10, 2020

View reviewed changes

lukeelmers merged commit 3b5ed46 into elastic:master Dec 10, 2020

lukeelmers deleted the feat/esaggs-args branch December 10, 2020 14:40

lukeelmers mentioned this pull request Dec 10, 2020

[7.x] [data.search] Clean up arguments to esaggs. (#84973) #85578

Merged

lukeelmers added a commit that referenced this pull request Dec 10, 2020

[7.x] [data.search] Clean up arguments to esaggs. (#84973) (#85578)

b16834b

wylieconlon mentioned this pull request Dec 14, 2020

[Lens] Use agg functions instead of string #77565

Closed

7 tasks

mbondyra mentioned this pull request Dec 16, 2020

[Lens] add typeToAgg for median #86078

Merged

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

[data.search] Clean up arguments to esaggs. #84973

[data.search] Clean up arguments to esaggs. #84973

lukeelmers commented Dec 4, 2020 •

edited

Loading

flash1293 Dec 4, 2020

lukeelmers Dec 4, 2020

lukeelmers Dec 7, 2020

lukeelmers Dec 7, 2020

lukeelmers Dec 7, 2020

lukeelmers Dec 7, 2020

lukeelmers Dec 7, 2020

lukeelmers Dec 7, 2020

lukeelmers Dec 7, 2020

stratoula Dec 10, 2020

lukeelmers Dec 7, 2020

lukeelmers Dec 7, 2020

lukeelmers Dec 7, 2020 •

edited

Loading

flash1293 Dec 8, 2020

elasticmachine commented Dec 7, 2020

lukeelmers Dec 9, 2020

lukeelmers Dec 9, 2020

lukeelmers Dec 9, 2020

lukeelmers Dec 9, 2020

lukeelmers Dec 9, 2020

ppisljar left a comment

flash1293 left a comment

flash1293 Dec 9, 2020

lukeelmers Dec 9, 2020

lukeelmers commented Dec 9, 2020

kibanamachine commented Dec 9, 2020

flash1293 left a comment

flash1293 commented Dec 10, 2020

stratoula left a comment

lukeelmers commented Dec 10, 2020

		let expectExpression: ExpectExpression;

		const expectClientToMatchServer = async (title: string, expression: string) => {

[data.search] Clean up arguments to esaggs. #84973

[data.search] Clean up arguments to esaggs. #84973

Conversation

lukeelmers commented Dec 4, 2020 • edited Loading

Summary

Testing

Plugin API Changes

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

lukeelmers Dec 7, 2020 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

elasticmachine commented Dec 7, 2020

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

ppisljar left a comment

Choose a reason for hiding this comment

flash1293 left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

lukeelmers commented Dec 9, 2020

kibanamachine commented Dec 9, 2020

💚 Build Succeeded

Metrics [docs]

Module Count

Async chunks

Distributable file count

Page load bundle

History

flash1293 left a comment

Choose a reason for hiding this comment

flash1293 commented Dec 10, 2020

stratoula left a comment

Choose a reason for hiding this comment

lukeelmers commented Dec 10, 2020

lukeelmers commented Dec 4, 2020 •

edited

Loading

lukeelmers Dec 7, 2020 •

edited

Loading