Skip to content

Latest commit

 

History

History
98 lines (70 loc) · 4.16 KB

applied-limits.md

File metadata and controls

98 lines (70 loc) · 4.16 KB

Applied limits for MLRun Quality Gate

MLRun Client

  1. The MLRun Client (in version >= 1.5.2) does not support OS Windows, see ML-4907. Know issues:

NOTE: Solution, it is necessity to use WSL2 under OS Windows

Feature Vector

  1. The data read (via feature vector) accept only ONE on-line and one off-line target in FeatureSet, see Slack discussion
    • in case of e.g. more on-line targets, it is not possible to choose relevant target for FeatureVector

RedisNoSqlTarget

  1. Issue with support date type

SQLTarget

NOTE: It is in preview version, very limited with focus on MySQL only, see detail below

  1. SQLTarget limits

    • missing support MORE primary keys (only ONE primary key is supported right now)
    • schema for mapping FeatureStore to Table must be defined manually (not automatically)
    • see detail mlrun/mlrun#5051
  2. SqlTarget is limited to MySql, if you need to create table (SqlTarget is in Technical Preview)

    • see the detail mlrun/mlrun#5231
    • NOTE: It is possible to use work-arround, create table before the ingest
  3. SqlTarget issue with save/load content mapping for SqlTarget

KafkaTarget

  1. It is not possible to use feature vector operations in KafkaTarget
    • but it is possible to do ingest to the KafkaTarget or consume data from KafkaSource (with triggers)
    • it means, test scenarios TS401, TS501, TS502, TS601 and TS701 for KafkaTarget was switch-off

KafkaSource

  1. It is necessery to ingest sample data to featureset, before kafka ingest
    • known issue from MLRun version v1.1.0, see ML-2407

HttpSource

  1. It is necessery to ingest sample data to featureset, before http ingest

CSVSource

  1. CSVSource supports only default CSV setting, it means sep=',', decimal='.'. It is not possible to use e.g. na_filter=True (it is the issue for nan value or empty strings).
    • in case of different setting, it is better to use Pandas/DataFrame source (it has bigger variability)

Pipelines

  1. The step DateExtractor, in case of part e.g. 'is_month_end', it generated the warning Converting input from bool to <class 'numpy.uint8'> for compatibility.

  2. Relation MapValues and storey.Filter see

  • Workarround: Use MapValues with parameter with_original_features=True

Ingest issues

  1. Clarification of issues with relation to the source/target and setting of InferOptions. See, full detail

Others

  1. Not to use the engine pandas

    • this pandas engine is useful only for test purpose (see the first info about that in change log for MLRun version 1.6.0)
  2. Issue with project delete

    • see mistake in MLRun versions 1.6.0, 1.6.1, 1.6.2, 1.6.3
    • expected solution in MLRun 1.7.0

Recommendation

  1. It is really tricky to use functions in Preview state
    • it makes sense to wait for full version delivery or contribute changes in Github