When instrumenting Kibana, aborted requests crash the server

[dgieselaar]

Kibana version:

7.6

Describe the bug:

Our cloud instance has crashed several times since instrumenting Kibana with the APM agent:

TypeError: request.abort is not a function
    at AbortSignal.options.signal.addEventListener (/home/apm/elastic/kibana/src/core/server/elasticsearch/cluster_client.ts:67:19)
    at Timeout._onTimeout (/home/apm/elastic/kibana/node_modules/abortcontroller-polyfill/dist/cjs-ponyfill.js:153:27)
    at ontimeout (timers.js:436:11)
    at tryOnTimeout (timers.js:300:5)
    at listOnTimeout (timers.js:263:5)
    at Timer.processTimers (timers.js:223:10)
Sending error to Elastic APM { id: '82c0025cbfbe1e4359e3da795e16758c' }

request is a promise returned by the Elasticsearch client. abort() will be called when for some reason the consumer wants to abort the request (for instance, when the request is taking too long). However, abort() is no longer available when the APM agent has instrumented Kibana, perhaps because it intercepts the promise but does not copy over the abort method.

Steps to reproduce:

1. Activate the APM agent by creating config/apm.dev.js:

module.exports = { active: true };

2. add the following lines to src/legacy/core_plugins/elasticsearch/lib/abortable_request_handler.js:

setTimeout(() => {
  console.log('aborting')
  controller.abort();
}, 100);

This will abort a request to the ES proxy endpoint (/elasticsearch/_search) after 100ms.

3. Then, go to Management > Index Patterns > Create index pattern. This should then call the ES proxy endpoint and crash Kibana.

Expected behavior:

Kibana should not crash when a request times out for whatever reason.

Any additional context:

The impact will be small, because users will not have instrumented Kibana with the APM agent, and it only happens on long-running requests or aborts.

@sqren @watson

[elasticmachine]

Pinging @elastic/kibana-platform (Team:Platform)

[pgayvallet]

@watson any progress on elastic/apm-agent-nodejs#1566?

[watson]

Unfortunately not, sorry. It's on my todo-list. Here's the PR's that I've made to fix it, but I got stuck on a failing test and never got any further. If anybody has time to help I'd be happy to review:

• fix: end span if outgoing http request ends prematurely apm-agent-nodejs#1583 - Status: One test is still failing on Node.js 8. I haven't had time to figure out why and fix it.
• fix(elasticsearch): ensure requests can be aborted apm-agent-nodejs#1566 - Status: Waiting for fix: end span if outgoing http request ends prematurely apm-agent-nodejs#1583 to merge.