Improving performance of Logstash Pipeline Viewer #27513
Labels
bug
Fixes for quality problems that affect the customer experience
Feature:Logstash Pipelines
Logstash Pipeline UI related
Team:Monitoring
Stack Monitoring team
Comments
ycombinator
added
Feature:Logstash Pipelines
|
Pinging @elastic/stack-monitoring |
ycombinator
added
the
bug
5 tasks
ycombinator
added a commit
that referenced
this issue
Mar 21, 2019
_This PR targets a feature branch instead of `master`. Once this PR is merged, a follow up PR will be made to the feature branch to update tests. The idea here is to keep the size of each PR small. Finally, a PR will be created from the feature branch to `master`._ Resolves #27513. ## Before this PR The Logstash Pipeline Viewer UI would make a single Kibana API call to fetch all the information necessary to render the Logstash pipeline. This included information necessary to render the detail drawer that opens up when a user clicks on an individual vertex in the pipeline. Naturally, this single API call fetched _a lot_ of data, not just from the Kibana server but also, in turn, from Elasticsearch as well. The "pro" of this approach was that the user would see instantaneous results if they clicked on a vertex in a pipeline and opened the detail drawer for that vertex. The "cons" were the amount of computation Elasticsearch had to perform and the amount of data being transferred over the wire between Elasticsearch and the Kibana server as well as between the Kibana server and the browser. ## With this PR This PR makes the Kibana API call to fetch data necessary for **initially** rendering the pipeline — that is, with the detail drawer closed — much lighter. When the user clicks on a vertex in a pipeline, a second API call is then made to fetch data necessary for the detail drawer. Concretely, this PR makes the following changes: * [x] For each metric of each vertex in the pipeline viewer, instead of returning a timeseries in the Kibana API response, we now only return the _latest_ data point. In other words, instead of returning `[ [t1, v1], [t2, v2], ... [tn, vn] ]` we now only return `vn`. This dramatically cuts down the number of aggregation buckets that Elasticsearch needs to generate. * [x] For each vertex in the pipeline viewer, instead of returning up to 9 metrics in the Kibana API response we now only return 4. These are the 4 that are actually needed for the initial rendering of the pipeline (i.e. with the detail drawer closed). The 5 metrics that are no longer returned are `events_in`, `events_out`, `duration_in_millis`, `queue_push_duration_in_millis`, and `queue_push_duration_in_millis_per_event`. * [x] Since we no longer return `queue_push_duration_in_millis`, and `queue_push_duration_in_millis_per_event` from the Kibana API response, we also no longer compute the aggregations behind these metrics. Specifically, we no longer need the `queue_push_duration_in_millis_temp_by_ephemeral_id` and `queue_push_duration_in_millis_total` aggregations. * [x] We no longer return the `pipeline.representation.plugins` array in the Kibana API response, since this array is not currently used in the Pipeline Viewer UI. We do this by excluding the `plugins` field from the Elasticsearch API response for the `logstash_state` document. * [x] A new API to return timeseries metrics for a single vertex is created. This API is called whenever the detail drawer is open and showing a plugin vertex. ## Gains, by the numbers Based on a simple, 1-input, 1-filter, and 1-output pipeline. * Before this PR, the Elasticsearch `logstash_stats` API responses (multiple calls were made using the `composite` aggregation over the `date_histogram` aggregation) generated a total of 1228 aggregation buckets (before any `filter_path`s were applied but across all `composite` "pages"). With this PR, the single `logstash_stats` API response (note that this is just for the initial rendering of the pipeline, with the detail drawer closed) generated 12 buckets (also before any `filter_path`s were applied). That's a **99.02% reduction** in number of buckets. * Before this PR, the Elasticsearch `logstash_stats` API responses added up to 70319 bytes. With this PR, the single `logstash_stats` API response for the same pipeline is 746 bytes. That's a **98.93% reduction** in size. * Before this PR, the Elasticsearch `logstash_state` API response was 7718 bytes. With this PR, the API response for the same pipeline is 2328 bytes. That's a **69.83% reduction** in size. * Before this PR the Kibana API response was 51777 bytes. With this PR, the API response for the same pipeline is 2567 bytes (again, note that this is just for the initial rendering of the pipeline, with the detail drawer closed). That's a **95.04% reduction** in size. ## Testing this PR 1. [Setup Logstash Monitoring](https://www.elastic.co/guide/en/logstash/current/configuring-logstash.html). 2. Start a Logstash pipeline, like so: ``` ./bin/logstash -e 'input { generator {} } filter { drop {} } output { stdout { codec => dots } }' ``` 3. Let the pipeline run for ~30 minutes (due to [this bug](#28153) 😞). 4. Open the Monitoring UI and visit the Logstash Pipeline Viewer for your pipeline. 5. Make sure the pipeline shows up. Check that the API response for the initial rendering of the pipeline doesn't contain any timeseries data. Then click on a vertex to open the Detail Drawer. Check that the details for that vertex show up, including sparklines rendered using timeseries data. Sample screenshots of what to expect: * Initial rendering <img width="1182" alt="Screen Shot 2019-03-18 at 11 02 45 AM" src="https://user-images.githubusercontent.com/51061/54552538-87caf980-496d-11e9-93ad-ede1eb6defa9.png"> * With detail drawer open for `drop` filter plugin vertex <img width="1201" alt="Screen Shot 2019-03-18 at 11 02 57 AM" src="https://user-images.githubusercontent.com/51061/54552548-8dc0da80-496d-11e9-9e9e-8d130ed361ca.png">
ycombinator
added a commit
that referenced
this issue
Mar 26, 2019
Resolves #27513. _This PR is a combination of #31293 (the code changes) + #33570 (test updates). These two PRs were individually reviewed and merged into a feature branch. This combo PR here simply sets up the merge from the feature branch to `master`._ Summary of changes, taken from #31293: ## Before this PR The Logstash Pipeline Viewer UI would make a single Kibana API call to fetch all the information necessary to render the Logstash pipeline. This included information necessary to render the detail drawer that opens up when a user clicks on an individual vertex in the pipeline. Naturally, this single API call fetched _a lot_ of data, not just from the Kibana server but also, in turn, from Elasticsearch as well. The "pro" of this approach was that the user would see instantaneous results if they clicked on a vertex in a pipeline and opened the detail drawer for that vertex. The "cons" were the amount of computation Elasticsearch had to perform and the amount of data being transferred over the wire between Elasticsearch and the Kibana server as well as between the Kibana server and the browser. ## With this PR This PR makes the Kibana API call to fetch data necessary for **initially** rendering the pipeline — that is, with the detail drawer closed — much lighter. When the user clicks on a vertex in a pipeline, a second API call is then made to fetch data necessary for the detail drawer. ## Gains, by the numbers Based on a simple, 1-input, 1-filter, and 1-output pipeline. * Before this PR, the Elasticsearch `logstash_stats` API responses (multiple calls were made using the `composite` aggregation over the `date_histogram` aggregation) generated a total of 1228 aggregation buckets (before any `filter_path`s were applied but across all `composite` "pages"). With this PR, the single `logstash_stats` API response (note that this is just for the initial rendering of the pipeline, with the detail drawer closed) generated 12 buckets (also before any `filter_path`s were applied). That's a **99.02% reduction** in number of buckets. * Before this PR, the Elasticsearch `logstash_stats` API responses added up to 70319 bytes. With this PR, the single `logstash_stats` API response for the same pipeline is 746 bytes. That's a **98.93% reduction** in size. * Before this PR, the Elasticsearch `logstash_state` API response was 7718 bytes. With this PR, the API response for the same pipeline is 2328 bytes. That's a **69.83% reduction** in size. * Before this PR the Kibana API response was 51777 bytes. With this PR, the API response for the same pipeline is 2567 bytes (again, note that this is just for the initial rendering of the pipeline, with the detail drawer closed). That's a **95.04% reduction** in size.
ycombinator
added a commit
to ycombinator/kibana
that referenced
this issue
Mar 26, 2019
Resolves elastic#27513. _This PR is a combination of elastic#31293 (the code changes) + elastic#33570 (test updates). These two PRs were individually reviewed and merged into a feature branch. This combo PR here simply sets up the merge from the feature branch to `master`._ Summary of changes, taken from elastic#31293: ## Before this PR The Logstash Pipeline Viewer UI would make a single Kibana API call to fetch all the information necessary to render the Logstash pipeline. This included information necessary to render the detail drawer that opens up when a user clicks on an individual vertex in the pipeline. Naturally, this single API call fetched _a lot_ of data, not just from the Kibana server but also, in turn, from Elasticsearch as well. The "pro" of this approach was that the user would see instantaneous results if they clicked on a vertex in a pipeline and opened the detail drawer for that vertex. The "cons" were the amount of computation Elasticsearch had to perform and the amount of data being transferred over the wire between Elasticsearch and the Kibana server as well as between the Kibana server and the browser. ## With this PR This PR makes the Kibana API call to fetch data necessary for **initially** rendering the pipeline — that is, with the detail drawer closed — much lighter. When the user clicks on a vertex in a pipeline, a second API call is then made to fetch data necessary for the detail drawer. ## Gains, by the numbers Based on a simple, 1-input, 1-filter, and 1-output pipeline. * Before this PR, the Elasticsearch `logstash_stats` API responses (multiple calls were made using the `composite` aggregation over the `date_histogram` aggregation) generated a total of 1228 aggregation buckets (before any `filter_path`s were applied but across all `composite` "pages"). With this PR, the single `logstash_stats` API response (note that this is just for the initial rendering of the pipeline, with the detail drawer closed) generated 12 buckets (also before any `filter_path`s were applied). That's a **99.02% reduction** in number of buckets. * Before this PR, the Elasticsearch `logstash_stats` API responses added up to 70319 bytes. With this PR, the single `logstash_stats` API response for the same pipeline is 746 bytes. That's a **98.93% reduction** in size. * Before this PR, the Elasticsearch `logstash_state` API response was 7718 bytes. With this PR, the API response for the same pipeline is 2328 bytes. That's a **69.83% reduction** in size. * Before this PR the Kibana API response was 51777 bytes. With this PR, the API response for the same pipeline is 2567 bytes (again, note that this is just for the initial rendering of the pipeline, with the detail drawer closed). That's a **95.04% reduction** in size.
ycombinator
added a commit
to ycombinator/kibana
that referenced
this issue
Mar 26, 2019
Resolves elastic#27513. _This PR is a combination of elastic#31293 (the code changes) + elastic#33570 (test updates). These two PRs were individually reviewed and merged into a feature branch. This combo PR here simply sets up the merge from the feature branch to `master`._ Summary of changes, taken from elastic#31293: The Logstash Pipeline Viewer UI would make a single Kibana API call to fetch all the information necessary to render the Logstash pipeline. This included information necessary to render the detail drawer that opens up when a user clicks on an individual vertex in the pipeline. Naturally, this single API call fetched _a lot_ of data, not just from the Kibana server but also, in turn, from Elasticsearch as well. The "pro" of this approach was that the user would see instantaneous results if they clicked on a vertex in a pipeline and opened the detail drawer for that vertex. The "cons" were the amount of computation Elasticsearch had to perform and the amount of data being transferred over the wire between Elasticsearch and the Kibana server as well as between the Kibana server and the browser. This PR makes the Kibana API call to fetch data necessary for **initially** rendering the pipeline — that is, with the detail drawer closed — much lighter. When the user clicks on a vertex in a pipeline, a second API call is then made to fetch data necessary for the detail drawer. Based on a simple, 1-input, 1-filter, and 1-output pipeline. * Before this PR, the Elasticsearch `logstash_stats` API responses (multiple calls were made using the `composite` aggregation over the `date_histogram` aggregation) generated a total of 1228 aggregation buckets (before any `filter_path`s were applied but across all `composite` "pages"). With this PR, the single `logstash_stats` API response (note that this is just for the initial rendering of the pipeline, with the detail drawer closed) generated 12 buckets (also before any `filter_path`s were applied). That's a **99.02% reduction** in number of buckets. * Before this PR, the Elasticsearch `logstash_stats` API responses added up to 70319 bytes. With this PR, the single `logstash_stats` API response for the same pipeline is 746 bytes. That's a **98.93% reduction** in size. * Before this PR, the Elasticsearch `logstash_state` API response was 7718 bytes. With this PR, the API response for the same pipeline is 2328 bytes. That's a **69.83% reduction** in size. * Before this PR the Kibana API response was 51777 bytes. With this PR, the API response for the same pipeline is 2567 bytes (again, note that this is just for the initial rendering of the pipeline, with the detail drawer closed). That's a **95.04% reduction** in size.
ycombinator
added a commit
to ycombinator/kibana
that referenced
this issue
Mar 26, 2019
Resolves elastic#27513. _This PR is a combination of elastic#31293 (the code changes) + elastic#33570 (test updates). These two PRs were individually reviewed and merged into a feature branch. This combo PR here simply sets up the merge from the feature branch to `master`._ Summary of changes, taken from elastic#31293: The Logstash Pipeline Viewer UI would make a single Kibana API call to fetch all the information necessary to render the Logstash pipeline. This included information necessary to render the detail drawer that opens up when a user clicks on an individual vertex in the pipeline. Naturally, this single API call fetched _a lot_ of data, not just from the Kibana server but also, in turn, from Elasticsearch as well. The "pro" of this approach was that the user would see instantaneous results if they clicked on a vertex in a pipeline and opened the detail drawer for that vertex. The "cons" were the amount of computation Elasticsearch had to perform and the amount of data being transferred over the wire between Elasticsearch and the Kibana server as well as between the Kibana server and the browser. This PR makes the Kibana API call to fetch data necessary for **initially** rendering the pipeline — that is, with the detail drawer closed — much lighter. When the user clicks on a vertex in a pipeline, a second API call is then made to fetch data necessary for the detail drawer. Based on a simple, 1-input, 1-filter, and 1-output pipeline. * Before this PR, the Elasticsearch `logstash_stats` API responses (multiple calls were made using the `composite` aggregation over the `date_histogram` aggregation) generated a total of 1228 aggregation buckets (before any `filter_path`s were applied but across all `composite` "pages"). With this PR, the single `logstash_stats` API response (note that this is just for the initial rendering of the pipeline, with the detail drawer closed) generated 12 buckets (also before any `filter_path`s were applied). That's a **99.02% reduction** in number of buckets. * Before this PR, the Elasticsearch `logstash_stats` API responses added up to 70319 bytes. With this PR, the single `logstash_stats` API response for the same pipeline is 746 bytes. That's a **98.93% reduction** in size. * Before this PR, the Elasticsearch `logstash_state` API response was 7718 bytes. With this PR, the API response for the same pipeline is 2328 bytes. That's a **69.83% reduction** in size. * Before this PR the Kibana API response was 51777 bytes. With this PR, the API response for the same pipeline is 2567 bytes (again, note that this is just for the initial rendering of the pipeline, with the detail drawer closed). That's a **95.04% reduction** in size.
ycombinator
added a commit
that referenced
this issue
Mar 26, 2019
Resolves #27513. _This PR is a combination of #31293 (the code changes) + #33570 (test updates). These two PRs were individually reviewed and merged into a feature branch. This combo PR here simply sets up the merge from the feature branch to `master`._ Summary of changes, taken from #31293: ## Before this PR The Logstash Pipeline Viewer UI would make a single Kibana API call to fetch all the information necessary to render the Logstash pipeline. This included information necessary to render the detail drawer that opens up when a user clicks on an individual vertex in the pipeline. Naturally, this single API call fetched _a lot_ of data, not just from the Kibana server but also, in turn, from Elasticsearch as well. The "pro" of this approach was that the user would see instantaneous results if they clicked on a vertex in a pipeline and opened the detail drawer for that vertex. The "cons" were the amount of computation Elasticsearch had to perform and the amount of data being transferred over the wire between Elasticsearch and the Kibana server as well as between the Kibana server and the browser. ## With this PR This PR makes the Kibana API call to fetch data necessary for **initially** rendering the pipeline — that is, with the detail drawer closed — much lighter. When the user clicks on a vertex in a pipeline, a second API call is then made to fetch data necessary for the detail drawer. ## Gains, by the numbers Based on a simple, 1-input, 1-filter, and 1-output pipeline. * Before this PR, the Elasticsearch `logstash_stats` API responses (multiple calls were made using the `composite` aggregation over the `date_histogram` aggregation) generated a total of 1228 aggregation buckets (before any `filter_path`s were applied but across all `composite` "pages"). With this PR, the single `logstash_stats` API response (note that this is just for the initial rendering of the pipeline, with the detail drawer closed) generated 12 buckets (also before any `filter_path`s were applied). That's a **99.02% reduction** in number of buckets. * Before this PR, the Elasticsearch `logstash_stats` API responses added up to 70319 bytes. With this PR, the single `logstash_stats` API response for the same pipeline is 746 bytes. That's a **98.93% reduction** in size. * Before this PR, the Elasticsearch `logstash_state` API response was 7718 bytes. With this PR, the API response for the same pipeline is 2328 bytes. That's a **69.83% reduction** in size. * Before this PR the Kibana API response was 51777 bytes. With this PR, the API response for the same pipeline is 2567 bytes (again, note that this is just for the initial rendering of the pipeline, with the detail drawer closed). That's a **95.04% reduction** in size.
ycombinator
added a commit
that referenced
this issue
Mar 26, 2019
Resolves #27513. _This PR is a combination of #31293 (the code changes) + #33570 (test updates). These two PRs were individually reviewed and merged into a feature branch. This combo PR here simply sets up the merge from the feature branch to `master`._ Summary of changes, taken from #31293: The Logstash Pipeline Viewer UI would make a single Kibana API call to fetch all the information necessary to render the Logstash pipeline. This included information necessary to render the detail drawer that opens up when a user clicks on an individual vertex in the pipeline. Naturally, this single API call fetched _a lot_ of data, not just from the Kibana server but also, in turn, from Elasticsearch as well. The "pro" of this approach was that the user would see instantaneous results if they clicked on a vertex in a pipeline and opened the detail drawer for that vertex. The "cons" were the amount of computation Elasticsearch had to perform and the amount of data being transferred over the wire between Elasticsearch and the Kibana server as well as between the Kibana server and the browser. This PR makes the Kibana API call to fetch data necessary for **initially** rendering the pipeline — that is, with the detail drawer closed — much lighter. When the user clicks on a vertex in a pipeline, a second API call is then made to fetch data necessary for the detail drawer. Based on a simple, 1-input, 1-filter, and 1-output pipeline. * Before this PR, the Elasticsearch `logstash_stats` API responses (multiple calls were made using the `composite` aggregation over the `date_histogram` aggregation) generated a total of 1228 aggregation buckets (before any `filter_path`s were applied but across all `composite` "pages"). With this PR, the single `logstash_stats` API response (note that this is just for the initial rendering of the pipeline, with the detail drawer closed) generated 12 buckets (also before any `filter_path`s were applied). That's a **99.02% reduction** in number of buckets. * Before this PR, the Elasticsearch `logstash_stats` API responses added up to 70319 bytes. With this PR, the single `logstash_stats` API response for the same pipeline is 746 bytes. That's a **98.93% reduction** in size. * Before this PR, the Elasticsearch `logstash_state` API response was 7718 bytes. With this PR, the API response for the same pipeline is 2328 bytes. That's a **69.83% reduction** in size. * Before this PR the Kibana API response was 51777 bytes. With this PR, the API response for the same pipeline is 2567 bytes (again, note that this is just for the initial rendering of the pipeline, with the detail drawer closed). That's a **95.04% reduction** in size.
ycombinator
added a commit
that referenced
this issue
Mar 26, 2019
Resolves #27513. _This PR is a combination of #31293 (the code changes) + #33570 (test updates). These two PRs were individually reviewed and merged into a feature branch. This combo PR here simply sets up the merge from the feature branch to `master`._ Summary of changes, taken from #31293: The Logstash Pipeline Viewer UI would make a single Kibana API call to fetch all the information necessary to render the Logstash pipeline. This included information necessary to render the detail drawer that opens up when a user clicks on an individual vertex in the pipeline. Naturally, this single API call fetched _a lot_ of data, not just from the Kibana server but also, in turn, from Elasticsearch as well. The "pro" of this approach was that the user would see instantaneous results if they clicked on a vertex in a pipeline and opened the detail drawer for that vertex. The "cons" were the amount of computation Elasticsearch had to perform and the amount of data being transferred over the wire between Elasticsearch and the Kibana server as well as between the Kibana server and the browser. This PR makes the Kibana API call to fetch data necessary for **initially** rendering the pipeline — that is, with the detail drawer closed — much lighter. When the user clicks on a vertex in a pipeline, a second API call is then made to fetch data necessary for the detail drawer. Based on a simple, 1-input, 1-filter, and 1-output pipeline. * Before this PR, the Elasticsearch `logstash_stats` API responses (multiple calls were made using the `composite` aggregation over the `date_histogram` aggregation) generated a total of 1228 aggregation buckets (before any `filter_path`s were applied but across all `composite` "pages"). With this PR, the single `logstash_stats` API response (note that this is just for the initial rendering of the pipeline, with the detail drawer closed) generated 12 buckets (also before any `filter_path`s were applied). That's a **99.02% reduction** in number of buckets. * Before this PR, the Elasticsearch `logstash_stats` API responses added up to 70319 bytes. With this PR, the single `logstash_stats` API response for the same pipeline is 746 bytes. That's a **98.93% reduction** in size. * Before this PR, the Elasticsearch `logstash_state` API response was 7718 bytes. With this PR, the API response for the same pipeline is 2328 bytes. That's a **69.83% reduction** in size. * Before this PR the Kibana API response was 51777 bytes. With this PR, the API response for the same pipeline is 2567 bytes (again, note that this is just for the initial rendering of the pipeline, with the detail drawer closed). That's a **95.04% reduction** in size.
mistic
added a commit
that referenced
this issue
Mar 26, 2019
* chore(NA): first changes on every package.json order to support new babel 7. chore(NA): build for kbn-pm with babel 7. * chore(NA): patch babel register to load typescrit * chore(NA): first working version with babel 7 replacing typescript compiler. * fix(NA): common preset declaration in order to make it work with babel-loader. * chore(na): organizing babel preset env package json. * chore(NA): mocha tests enabled. * fix(NA): typo on importing * test(NA): majority of x-pack tests ported to use babel-jest * fix(NA): report info button test with babel-jest. * fix(NA): polling service tests. * test(na): fix server plugins plugin tests. * test(NA): batch of test fixs for jest tests under babel-jest hoisting. * chore(NA): add babel plugin to hoist mock prefixed vars on jest tests. * chore(NA): update yarn.lock file. * chore(NA): tests passing. * chore(NA): remove wrong dep * chore(NA): fix tsconfig * chore(NA): skip babel for ts-jest. * chore(NA): selectively apply the plugin to strip off namespace from ts files. * chore(NA): remove not needed changes from ts tests * chore(NA): removed ts-jest dependency. chore(NA): migrate ts tests on x-pack to use babel-jest with the new pattern. * chore(NA): migrate kibana default distribution typescript tests to run with babel-jest and the new test mock pattern. * chore(NA): merge and solve conflicts with master. * chore(NA): fix problems reported by eslint * chore(NA): fix license ovveride for babel-plugin-mock-imports * chore(NA): update jest integration tests for kbn pm * chore(NA): update babel jest integration tests for kbn pm. * test(NA): update jest integration snapshot for kbn pm. * chore(NA): apply changes according to the pull request reviews. * chore(NA): apply changes according to the pull request reviews. * refact(NA): migrate jest tests to the new pattern. * fix(NA): babel 7 polyfill in the tests bundle. * chore(NA): restore needed step in order to compile x-pack with typescript. * chore(NA): change build to compile typescript with babel for the oss code. chore(NA): change transpile typescript task to only transpile types for x-pack. refact(NA): common preset for babel 7 * Revert "chore(NA): change build to compile typescript with babel for the oss code. chore(NA): change transpile typescript task to only transpile types for x-pack. refact(NA): common preset for babel 7" This reverts commit 2707d53. * fix(NA): import paths for tabConfigConst * chore(NA): fix transpiling error on browser tests * chore(NA): simplify kbn babel preset package. * chore(NA): migrate build to use babel transpiler for typescript excluding xpack. * fix(NA): introduced error on test quick task. * fix(NA): fix preset for client side code on build. * fix(NA): build with babel * fix(NA): negated patterns in the end. * fix(NA): kbn_tp_sample_panel_action creation. * fix(NA): babel typescript transform plugin workaround when exporting interface name. * refact(NA): remove not needed type cast to any on jest test. * docs(NA): add developement documentation about jest mocks test pattern. * chore(NA): missing unmerged path. * chore(NA): fix jest tests for template. * [CCR] Client integration tests (table lists) (#33525) * Force user to re-authenticate if token refresh fails with `400` status code. (#33774) * Improve performance of the Logstash Pipeline Viewer (#33793) Resolves #27513. _This PR is a combination of #31293 (the code changes) + #33570 (test updates). These two PRs were individually reviewed and merged into a feature branch. This combo PR here simply sets up the merge from the feature branch to `master`._ Summary of changes, taken from #31293: ## Before this PR The Logstash Pipeline Viewer UI would make a single Kibana API call to fetch all the information necessary to render the Logstash pipeline. This included information necessary to render the detail drawer that opens up when a user clicks on an individual vertex in the pipeline. Naturally, this single API call fetched _a lot_ of data, not just from the Kibana server but also, in turn, from Elasticsearch as well. The "pro" of this approach was that the user would see instantaneous results if they clicked on a vertex in a pipeline and opened the detail drawer for that vertex. The "cons" were the amount of computation Elasticsearch had to perform and the amount of data being transferred over the wire between Elasticsearch and the Kibana server as well as between the Kibana server and the browser. ## With this PR This PR makes the Kibana API call to fetch data necessary for **initially** rendering the pipeline — that is, with the detail drawer closed — much lighter. When the user clicks on a vertex in a pipeline, a second API call is then made to fetch data necessary for the detail drawer. ## Gains, by the numbers Based on a simple, 1-input, 1-filter, and 1-output pipeline. * Before this PR, the Elasticsearch `logstash_stats` API responses (multiple calls were made using the `composite` aggregation over the `date_histogram` aggregation) generated a total of 1228 aggregation buckets (before any `filter_path`s were applied but across all `composite` "pages"). With this PR, the single `logstash_stats` API response (note that this is just for the initial rendering of the pipeline, with the detail drawer closed) generated 12 buckets (also before any `filter_path`s were applied). That's a **99.02% reduction** in number of buckets. * Before this PR, the Elasticsearch `logstash_stats` API responses added up to 70319 bytes. With this PR, the single `logstash_stats` API response for the same pipeline is 746 bytes. That's a **98.93% reduction** in size. * Before this PR, the Elasticsearch `logstash_state` API response was 7718 bytes. With this PR, the API response for the same pipeline is 2328 bytes. That's a **69.83% reduction** in size. * Before this PR the Kibana API response was 51777 bytes. With this PR, the API response for the same pipeline is 2567 bytes (again, note that this is just for the initial rendering of the pipeline, with the detail drawer closed). That's a **95.04% reduction** in size. * [Maps] split settings into layer and source panels (#33788) * [Maps] split settings into layer and source panels * fix SCSS import * [env] exit if starting as root (#21563) * [env] exit if starting as root * fix windows * s/--allow-root * Typescript sample panel action (#33602) * Typescript sample panel action * Update EUI version to match main cabana version * update yarn.lock * add back typings include * use correct relative path * Home page "recent links" should communicate saved object type #21896 (#33694) * adds object type for screen order * adds object type for pointer hovering * Update src/legacy/ui/public/chrome/directives/header_global_nav/components/header.tsx Co-Authored-By: rockfield <[email protected]>
joelgriffith
pushed a commit
that referenced
this issue
Mar 27, 2019
Resolves #27513. _This PR is a combination of #31293 (the code changes) + #33570 (test updates). These two PRs were individually reviewed and merged into a feature branch. This combo PR here simply sets up the merge from the feature branch to `master`._ Summary of changes, taken from #31293: ## Before this PR The Logstash Pipeline Viewer UI would make a single Kibana API call to fetch all the information necessary to render the Logstash pipeline. This included information necessary to render the detail drawer that opens up when a user clicks on an individual vertex in the pipeline. Naturally, this single API call fetched _a lot_ of data, not just from the Kibana server but also, in turn, from Elasticsearch as well. The "pro" of this approach was that the user would see instantaneous results if they clicked on a vertex in a pipeline and opened the detail drawer for that vertex. The "cons" were the amount of computation Elasticsearch had to perform and the amount of data being transferred over the wire between Elasticsearch and the Kibana server as well as between the Kibana server and the browser. ## With this PR This PR makes the Kibana API call to fetch data necessary for **initially** rendering the pipeline — that is, with the detail drawer closed — much lighter. When the user clicks on a vertex in a pipeline, a second API call is then made to fetch data necessary for the detail drawer. ## Gains, by the numbers Based on a simple, 1-input, 1-filter, and 1-output pipeline. * Before this PR, the Elasticsearch `logstash_stats` API responses (multiple calls were made using the `composite` aggregation over the `date_histogram` aggregation) generated a total of 1228 aggregation buckets (before any `filter_path`s were applied but across all `composite` "pages"). With this PR, the single `logstash_stats` API response (note that this is just for the initial rendering of the pipeline, with the detail drawer closed) generated 12 buckets (also before any `filter_path`s were applied). That's a **99.02% reduction** in number of buckets. * Before this PR, the Elasticsearch `logstash_stats` API responses added up to 70319 bytes. With this PR, the single `logstash_stats` API response for the same pipeline is 746 bytes. That's a **98.93% reduction** in size. * Before this PR, the Elasticsearch `logstash_state` API response was 7718 bytes. With this PR, the API response for the same pipeline is 2328 bytes. That's a **69.83% reduction** in size. * Before this PR the Kibana API response was 51777 bytes. With this PR, the API response for the same pipeline is 2567 bytes (again, note that this is just for the initial rendering of the pipeline, with the detail drawer closed). That's a **95.04% reduction** in size.
joelgriffith
pushed a commit
that referenced
this issue
Mar 27, 2019
* chore(NA): first changes on every package.json order to support new babel 7. chore(NA): build for kbn-pm with babel 7. * chore(NA): patch babel register to load typescrit * chore(NA): first working version with babel 7 replacing typescript compiler. * fix(NA): common preset declaration in order to make it work with babel-loader. * chore(na): organizing babel preset env package json. * chore(NA): mocha tests enabled. * fix(NA): typo on importing * test(NA): majority of x-pack tests ported to use babel-jest * fix(NA): report info button test with babel-jest. * fix(NA): polling service tests. * test(na): fix server plugins plugin tests. * test(NA): batch of test fixs for jest tests under babel-jest hoisting. * chore(NA): add babel plugin to hoist mock prefixed vars on jest tests. * chore(NA): update yarn.lock file. * chore(NA): tests passing. * chore(NA): remove wrong dep * chore(NA): fix tsconfig * chore(NA): skip babel for ts-jest. * chore(NA): selectively apply the plugin to strip off namespace from ts files. * chore(NA): remove not needed changes from ts tests * chore(NA): removed ts-jest dependency. chore(NA): migrate ts tests on x-pack to use babel-jest with the new pattern. * chore(NA): migrate kibana default distribution typescript tests to run with babel-jest and the new test mock pattern. * chore(NA): merge and solve conflicts with master. * chore(NA): fix problems reported by eslint * chore(NA): fix license ovveride for babel-plugin-mock-imports * chore(NA): update jest integration tests for kbn pm * chore(NA): update babel jest integration tests for kbn pm. * test(NA): update jest integration snapshot for kbn pm. * chore(NA): apply changes according to the pull request reviews. * chore(NA): apply changes according to the pull request reviews. * refact(NA): migrate jest tests to the new pattern. * fix(NA): babel 7 polyfill in the tests bundle. * chore(NA): restore needed step in order to compile x-pack with typescript. * chore(NA): change build to compile typescript with babel for the oss code. chore(NA): change transpile typescript task to only transpile types for x-pack. refact(NA): common preset for babel 7 * Revert "chore(NA): change build to compile typescript with babel for the oss code. chore(NA): change transpile typescript task to only transpile types for x-pack. refact(NA): common preset for babel 7" This reverts commit 2707d53. * fix(NA): import paths for tabConfigConst * chore(NA): fix transpiling error on browser tests * chore(NA): simplify kbn babel preset package. * chore(NA): migrate build to use babel transpiler for typescript excluding xpack. * fix(NA): introduced error on test quick task. * fix(NA): fix preset for client side code on build. * fix(NA): build with babel * fix(NA): negated patterns in the end. * fix(NA): kbn_tp_sample_panel_action creation. * fix(NA): babel typescript transform plugin workaround when exporting interface name. * refact(NA): remove not needed type cast to any on jest test. * docs(NA): add developement documentation about jest mocks test pattern. * chore(NA): missing unmerged path. * chore(NA): fix jest tests for template. * [CCR] Client integration tests (table lists) (#33525) * Force user to re-authenticate if token refresh fails with `400` status code. (#33774) * Improve performance of the Logstash Pipeline Viewer (#33793) Resolves #27513. _This PR is a combination of #31293 (the code changes) + #33570 (test updates). These two PRs were individually reviewed and merged into a feature branch. This combo PR here simply sets up the merge from the feature branch to `master`._ Summary of changes, taken from #31293: ## Before this PR The Logstash Pipeline Viewer UI would make a single Kibana API call to fetch all the information necessary to render the Logstash pipeline. This included information necessary to render the detail drawer that opens up when a user clicks on an individual vertex in the pipeline. Naturally, this single API call fetched _a lot_ of data, not just from the Kibana server but also, in turn, from Elasticsearch as well. The "pro" of this approach was that the user would see instantaneous results if they clicked on a vertex in a pipeline and opened the detail drawer for that vertex. The "cons" were the amount of computation Elasticsearch had to perform and the amount of data being transferred over the wire between Elasticsearch and the Kibana server as well as between the Kibana server and the browser. ## With this PR This PR makes the Kibana API call to fetch data necessary for **initially** rendering the pipeline — that is, with the detail drawer closed — much lighter. When the user clicks on a vertex in a pipeline, a second API call is then made to fetch data necessary for the detail drawer. ## Gains, by the numbers Based on a simple, 1-input, 1-filter, and 1-output pipeline. * Before this PR, the Elasticsearch `logstash_stats` API responses (multiple calls were made using the `composite` aggregation over the `date_histogram` aggregation) generated a total of 1228 aggregation buckets (before any `filter_path`s were applied but across all `composite` "pages"). With this PR, the single `logstash_stats` API response (note that this is just for the initial rendering of the pipeline, with the detail drawer closed) generated 12 buckets (also before any `filter_path`s were applied). That's a **99.02% reduction** in number of buckets. * Before this PR, the Elasticsearch `logstash_stats` API responses added up to 70319 bytes. With this PR, the single `logstash_stats` API response for the same pipeline is 746 bytes. That's a **98.93% reduction** in size. * Before this PR, the Elasticsearch `logstash_state` API response was 7718 bytes. With this PR, the API response for the same pipeline is 2328 bytes. That's a **69.83% reduction** in size. * Before this PR the Kibana API response was 51777 bytes. With this PR, the API response for the same pipeline is 2567 bytes (again, note that this is just for the initial rendering of the pipeline, with the detail drawer closed). That's a **95.04% reduction** in size. * [Maps] split settings into layer and source panels (#33788) * [Maps] split settings into layer and source panels * fix SCSS import * [env] exit if starting as root (#21563) * [env] exit if starting as root * fix windows * s/--allow-root * Typescript sample panel action (#33602) * Typescript sample panel action * Update EUI version to match main cabana version * update yarn.lock * add back typings include * use correct relative path * Home page "recent links" should communicate saved object type #21896 (#33694) * adds object type for screen order * adds object type for pointer hovering * Update src/legacy/ui/public/chrome/directives/header_global_nav/components/header.tsx Co-Authored-By: rockfield <[email protected]>
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
There have been cases reported of relatively small Logstash pipelines taking a long time to load, or simply not loading up at all, in the Pipeline Viewer Monitoring UI.
An easy way to test this would be to create a test Logstash pipeline containing ~30 vertices, including a few
ifvertices (so we have > ~30 edges). Configure Logstash Monitoring and let the pipeline run for several hours (I'll update the issue with more concrete numbers based on my testing). Then open the Pipeline Viewer Monitoring UI for the pipeline.The Kibana API that the Pipeline Viewer calls for its data returns the entire structure of the pipeline (vertices + edges, see the
pipeline.representation.graphproperty in the API response). Each vertex object has astatsproperty which is an object representing stats about that vertex. There can be up to 7 stats for each vertex. Each stat object has adataproperty which contains timeseries data for that stat (an array of timestamp+metric pairs).All in all, this generates a LOT of data points. A pipeline with 30 vertices would generate 30 x 7 x (# of timeseries buckets) data points. The # of timeseries buckets is determined by how long the pipeline version being viewed has been running. For instance, if it has been running for 24 hours, that would set the
date_histogrambucket size to 10 minutes, resulting in 144 timeseries buckets (24 * 60 / 10 = 144). So in the 30-vertex example, that would result in a total of 30 x 7 x 144 = 30,240 data points!However, if you look at the pipeline's rendering in the Pipeline Viewer, you'll only see 3 stats are shown per vertex. The remaining stats aren't visible but they are used for calculations that could affect the design of the vertex (e.g. highlighting it if the plugin represented by the vertex is consuming an unusually high amount of CPU).
Also, only the latest data point from the timeseries data for each of the 7 stats is considered for calculations or display. So why do we even get a timeseries from ES? Why not just get the latest data point? That's because if you click on a vertex, a vertex detail flyout opens up to the right. In this flyout, the timeseries for some of the stats is rendered as sparklines.
I suspect the performance issues caused by this page are related to large number of data points that the ES aggregation query has to generate (but we should confirm this with testing). If this is indeed the case, here might be one way to mitigate the issue:
For the initial rendering of the pipeline (i.e. with the detail flyout closed), only request the latest data points for each stat instead of a timeseries for each stat from ES (as opposed to today, where we request the timeseries for each stat, then pull out the latest data point from the timneseries in browser-side code). Then, when the user clicks on a particular vertex, request the timeseries for each stats for just that vertex, in order to render the sparklines in the detail flyout.
The text was updated successfully, but these errors were encountered: