I want to know whether the nanos length returned by the numberToHrtime method is a fixed length?

[yuanman0109]

opentelemetry-js/packages/opentelemetry-core/src/common/time.ts

Line 36 in 369b07e

function numberToHrtime(epochMillis: number): api.HrTime {

I saw this method used in many places, so I tested this method in the development environment. I found that sometimes the length of time obtained is unstable. Sometimes the length of the value 'nano' is 9, and sometimes it is 8. Is this normal? My Chrome version is the latest, and this is my screenshot：

[yuanman0109]

opentelemetry-js/packages/opentelemetry-core/src/common/time.ts

Line 140 in 369b07e

export function hrTimeToNanoseconds(time: api.HrTime): number {

The start and end time of the span are calculated by this method. If the precision of js will affect the calculation results here, our duration value will be very inaccurate, In about 1 second, the performance statistics will be affected @legendecas @dyladan

[legendecas]

Duplicate of #2643 for hrTimeToNanoseconds.

[legendecas]

numberToHrtime converts number of seconds to HrTime tuple ([seconds, nanoseconds]) without precision loss. Count of digits in the elements of the HrTime tuple can change depends on the leading zero counts of the element value.

[Flarna]

A bit strange is that nanos have a fractional part. I have never seen this in node.js process.hrtime().

Also doing a conversion via a string might be not the fastest variant and looks a likely a bit misleading.

[yuanman0109]

const NANOSECOND_DIGITS = 6;
const SECOND_TO_NANOSECONDS = Math.pow(10, NANOSECOND_DIGITS);
function numberToHrtime(epochMillis) {
  const epochSeconds = epochMillis / 1000;
  // Decimals only.
  const seconds = Math.trunc(epochSeconds);
  const time = seconds * 1000;
  // Round sub-nanosecond accuracy to nanosecond.
  const nanos = (epochMillis - time) * SECOND_TO_NANOSECONDS;
  return [seconds, nanos];
}

Avoid using decimals to temporarily solve this problem

[dyladan]

@legendecas should this be closed as a dup of #2643 ?

[legendecas]

Sorry, I might get it wrong in the above statement on numberToHrtime. I'll investigate the accuracy of the conversion of numberToHrtime with the fraction part (e.g. chrome performance.now() returns a number with the fractional part).