The neon_utils extension

The neon_utils extension provides a num_cpus() function you can use to monitor how Neon's Autoscaling feature allocates vCPU in response to workload. The function returns the current number of allocated vCPUs.

For information about Neon's Autoscaling feature, see Autoscaling.

Install the neon_utils extension

Install the neon_utils extension by running the following CREATE EXTENSION statement in the Neon SQL Editor or from a client such as psql that is connected to Neon.

CREATE EXTENSION neon_utils;

For information about using the Neon SQL Editor, see Query with Neon's SQL Editor. For information about using the psql client with Neon, see Connect with psql.

Use the num_cpus() function

In Neon, computing capacity is measured in Compute Units (CU). One CU is 1 vCPU and 4 GB of RAM, 2 CU is 2 vCPU and 8 GB of RAM, and so on. The amount of RAM in GB is always 4 times the number of vCPU. A Neon compute can have anywhere from .25 to 56 CU, but Autoscaling is only supported up to 16 CU.

Defining a minimum and maximum compute size for your compute, as shown below, enables autoscaling.

As your workload changes, computing capacity scales dynamically between the minimum and maximum settings defined in your compute configuration. To retrieve the number of allocated vCPU at any point in time, you can run the following query:

SELECT num_cpus();

For autoscaling configuration instructions, see Compute size and autoscaling configuration.

Limitations

The following limitations apply:

• The num_cpus() function does not return fractional vCPU sizes. The Autoscaling feature can scale by fractional vCPU, but the num_cpus() function reports the next whole number. For example, if the current number of allocated vCPU is .25 or .5, the num_cpus() function returns 1.
• The num_cpus() function only works on computes that have the Autoscaling feature enabled. Running the function on a fixed-size compute does not return a correct value.

Observe autoscaling with neon_utils and pgbench

The following instructions demonstrate how you can use the num_cpus() function with pgbench to observe how Neon's Autoscaling feature responds to workload.

Prerequisites

• Ensure that autoscaling is enabled for your compute. For instructions, see Compute size and autoscaling configuration. The following example uses a minimum setting of 0.25 Compute Units (CU) and a maximum of 4.
• The pgbench utility.

Run the test

1. Install the neon_utils extension:

   CREATE EXTENSION IF NOT EXISTS neon_utils;

2. Create a test.sql file with the following queries:

   SELECT LOG(factorial(5000)) / LOG(factorial(2500));
   SELECT txid_current();

3. To avoid errors when running pgbench, initialize your database with the tables used by pgbench. This can be done using the pgbench -i command, specifying the connection string for your Neon database. You can obtain a connection string from the Connection Details widget on the Neon Dashboard.

   pgbench -i postgresql://[user]:[password]@[neon_hostname]/[dbname]

4. Run a pgbench test with your test.sql file, specifying your connection string:

   pgbench -f test.sql -c 15 -T 1000 -P 1 postgresql://[user]:[password]@[neon_hostname]/[dbname]

   The test produces output similar to the following on a compute set to scale from 0.25 to 4 CUs.

   pgbench (15.3)
   starting vacuum...end.
   progress: 8.4 s, 0.0 tps, lat 0.000 ms stddev 0.000, 0 failed
   progress: 9.0 s, 0.0 tps, lat 0.000 ms stddev 0.000, 0 failed
   progress: 10.0 s, 4.0 tps, lat 1246.290 ms stddev 3.253, 0 failed
   progress: 11.0 s, 6.0 tps, lat 1892.455 ms stddev 446.686, 0 failed
   progress: 12.0 s, 9.0 tps, lat 2091.352 ms stddev 1068.303, 0 failed
   progress: 13.0 s, 5.0 tps, lat 1881.682 ms stddev 700.852, 0 failed
   progress: 14.0 s, 6.0 tps, lat 2660.009 ms stddev 1404.672, 0 failed
   progress: 15.0 s, 9.0 tps, lat 2354.776 ms stddev 1248.686, 0 failed
   progress: 16.0 s, 8.0 tps, lat 1770.870 ms stddev 776.465, 0 failed
   progress: 17.0 s, 7.0 tps, lat 1800.686 ms stddev 611.749, 0 failed
   progress: 18.0 s, 18.0 tps, lat 1681.841 ms stddev 1187.918, 0 failed
   progress: 19.0 s, 29.0 tps, lat 561.201 ms stddev 139.565, 0 failed
   progress: 20.0 s, 27.0 tps, lat 507.782 ms stddev 153.889, 0 failed
   progress: 21.0 s, 30.0 tps, lat 493.312 ms stddev 121.688, 0 failed
   progress: 22.0 s, 32.0 tps, lat 513.444 ms stddev 185.033, 0 failed
   progress: 23.0 s, 32.0 tps, lat 503.135 ms stddev 199.435, 0 failed
   progress: 24.0 s, 28.0 tps, lat 492.913 ms stddev 124.019, 0 failed
   progress: 25.0 s, 43.0 tps, lat 366.719 ms stddev 123.547, 0 failed
   progress: 26.0 s, 49.0 tps, lat 334.276 ms stddev 79.043, 0 failed
   progress: 27.0 s, 40.0 tps, lat 354.922 ms stddev 83.560, 0 failed
   progress: 28.0 s, 31.0 tps, lat 400.645 ms stddev 29.236, 0 failed
   progress: 29.0 s, 48.0 tps, lat 373.522 ms stddev 64.446, 0 failed
   progress: 30.0 s, 44.0 tps, lat 333.343 ms stddev 86.497, 0 failed
   progress: 31.0 s, 44.0 tps, lat 326.754 ms stddev 82.990, 0 failed
   progress: 32.0 s, 44.0 tps, lat 329.317 ms stddev 76.728, 0 failed
   progress: 33.0 s, 53.0 tps, lat 321.572 ms stddev 76.427, 0 failed
   progress: 34.0 s, 57.0 tps, lat 254.500 ms stddev 33.013, 0 failed
   progress: 35.0 s, 60.0 tps, lat 251.035 ms stddev 37.574, 0 failed
   progress: 36.0 s, 58.0 tps, lat 256.846 ms stddev 36.390, 0 failed
   progress: 37.0 s, 60.0 tps, lat 249.165 ms stddev 36.764, 0 failed
   progress: 38.0 s, 57.0 tps, lat 263.885 ms stddev 31.351, 0 failed
   progress: 39.0 s, 56.0 tps, lat 262.529 ms stddev 43.900, 0 failed
   progress: 40.0 s, 58.0 tps, lat 259.052 ms stddev 39.737, 0 failed
   ...

5. Call the num_cpus() function to retrieve the current number of allocated vCPU.

   ​​neondb=> SELECT num_cpus();
   num_cpus
   ----------
           4
   (1 row)