菜单
Patterns for reading and writing oleander Iceberg catalog tables in Spark jobs, including naming conventions, write modes, and catalog hierarchy.
默认使用会先检查来源的 Prompt；你也可以切换为直接命令，或下载本地副本。
决定是否安装前，请先阅读 SKILL.md，以及 SkillsMP 当前展示的配套文件。
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
直接命令不会经过审查 Prompt；运行前请先检查来源。
npx skills add https://github.com/OleanderHQ/pyspark-template --skill oleander-iceberg-catalog命令会保持在同一行。复制前请横向滚动并检查完整内容。
想先保存到本地？可下载 SkillsMP 当前能够提供的文件。
基于 SOC 职业分类
正在显示 SKILL.md
| name | oleander-iceberg-catalog |
| description | Patterns for reading and writing oleander Iceberg catalog tables in Spark jobs, including naming conventions, write modes, and catalog hierarchy. |
Use this skill when reading from or writing to the oleander Iceberg catalog in a Spark job.
Tables are addressed as catalog.namespace.table:
oleander for the managed Lakekeeper-backed catalogdefault, san_francisco, my_org)By default, default and telemetry namespaces are available out of the box. The telemetry namespace contains oleander-managed data you can query directly (for example oleander.telemetry.run_events, oleander.telemetry.traces, and oleander.telemetry.logs).
Examples:
oleander.default.sf_311
oleander.san_francisco.district_stats
oleander.my_namespace.results
Use spark.table() with the fully qualified name:
df = spark.table("oleander.default.sf_311")
Never construct table paths as raw S3 URIs. Always use the catalog name so Iceberg metadata and lineage are tracked correctly.
Append (add rows to an existing or new table):
df.writeTo("oleander.my_namespace.my_table").append()
Overwrite (replace table contents):
df.write.mode("overwrite").saveAsTable("oleander.my_namespace.my_table")
Use writeTo(...).append() for incremental pipelines. Use write.mode("overwrite").saveAsTable(...) when replacing full result sets each run.
Avoid collecting data to the driver and then writing from Python memory. Keep writes as Spark DataFrame operations so Iceberg handles the transaction, partitioning, and metadata correctly.
Bad:
rows = df.collect()
# write rows from Python memory
Good:
df.write.mode("overwrite").saveAsTable("oleander.my_namespace.my_table")
Accept table names as arguments or environment variables so scripts are reusable:
import os, argparse
parser = argparse.ArgumentParser()
parser.add_argument("--input-table", default="oleander.default.sf_311")
parser.add_argument("--output-catalog", default="oleander.my_namespace")
args = parser.parse_args()
df = spark.table(args.input_table)
df.write.mode("overwrite").saveAsTable(f"{args.output_catalog}.results")
If a table is read and used in multiple downstream transforms, cache it once and unpersist when done:
df = spark.table("oleander.default.sf_311")
df.cache()
# ... multiple transforms ...
df.unpersist()
Do not cache tables that are only used once.