Memory footprint excessive

[tback]

Environment details

• OS type and version: MacOS 15.1.1 (24B91)
• Python version: Python 3.12.0
• pip version: pip 24.3.1
• google-cloud-aiplatform version: Version: 1.72.0

Steps to reproduce

Get the memory footprint of python doing a simple and similar import. An unladen swallow is about 10MiB:

/usr/bin/time -l python -c 'import google.cloud'
        0.03 real         0.01 user         0.00 sys
            10878976  maximum resident set size   <-- 11Mi
                   0  average shared memory size
                   0  average unshared data size
                   0  average unshared stack size
                 906  page reclaims
                  14  page faults
                   0  swaps
                   0  block input operations
                   0  block output operations
                   0  messages sent
                   0  messages received
                   0  signals received
                  24  voluntary context switches
                  29  involuntary context switches
           134544684  instructions retired
            52246608  cycles elapsed
             7226432  peak memory footprint  <-- 6.9Mi

Get memory footprint when importing aiplatform. It uses around 180MiB just doing the basic import.

/usr/bin/time -l python -c 'import google.cloud.aiplatform'
        1.36 real         0.95 user         0.15 sys
           191102976  maximum resident set size  <-- 183Mi
                   0  average shared memory size
                   0  average unshared data size
                   0  average unshared stack size
               20349  page reclaims
                 381  page faults
                   0  swaps
                   0  block input operations
                   0  block output operations
                   0  messages sent
                   0  messages received
                   0  signals received
                2297  voluntary context switches
                 376  involuntary context switches
          8019160261  instructions retired
          3445365715  cycles elapsed
           178669120  peak memory footprint  <-- 171Mi

[tback]

Any news on this?

[davidemerritt]

We just ran into this as well, makes this unusable in a production API environment. Has this been triaged?

[niehongxu116]

the same problem..