Imported from jeremylongshore/tons-of-skills-marketplace (skills/.curated/palantir-performance-tuning/SKILL.md). Install upstream with npx skills add jeremylongshore/tons-of-skills-marketplace --skill palantir-performance-tuning. Copyright stays with the author (MIT).

Palantir Performance Tuning

Overview

Optimize Foundry API performance: efficient pagination, client-side caching, batch object retrieval, and Spark transform tuning with @configure profiles.

Prerequisites

• Completed palantir-install-auth setup
• Working Foundry integration to optimize
• Access to Foundry build metrics (for transform tuning)

Instructions

Step 1: Efficient Pagination

from functools import lru_cache

def fetch_all_objects(client, ontology: str, object_type: str, page_size: int = 500):
    """Fetch all objects with maximum page size to minimize API calls."""
    all_objects = []
    page_token = None
    while True:
        result = client.ontologies.OntologyObject.list(
            ontology=ontology,
            object_type=object_type,
            page_size=min(page_size, 500),  # Foundry max is 500
            page_token=page_token,
        )
        all_objects.extend(result.data)
        page_token = result.next_page_token
        if not page_token:
            break
    return all_objects

Step 2: Client-Side Caching

from cachetools import TTLCache
import hashlib, json

_cache = TTLCache(maxsize=1000, ttl=300)  # 5-minute TTL

def cached_get_object(client, ontology, object_type, primary_key):
    """Cache Ontology object reads to reduce API calls."""
    cache_key = f"{ontology}:{object_type}:{primary_key}"
    if cache_key in _cache:
        return _cache[cache_key]
    obj = client.ontologies.OntologyObject.get(
        ontology=ontology, object_type=object_type, primary_key=primary_key,
    )
    _cache[cache_key] = obj
    return obj

def invalidate_cache(ontology, object_type, primary_key):
    cache_key = f"{ontology}:{object_type}:{primary_key}"
    _cache.pop(cache_key, None)

Step 3: Batch Object Retrieval

def batch_get_objects(client, ontology, object_type, primary_keys, batch_size=50):
    """Retrieve multiple objects using search filter instead of individual GETs."""
    results = {}
    for i in range(0, len(primary_keys), batch_size):
        batch = primary_keys[i:i+batch_size]
        search_result = client.ontologies.OntologyObject.search(
            ontology=ontology,
            object_type=object_type,
            where={
                "type": "in",
                "field": "primaryKey",
                "value": batch,
            },
            page_size=batch_size,
        )
        for obj in search_result.data:
            pk = obj.properties.get("primaryKey", obj.rid)
            results[pk] = obj
    return results

Step 4: Transform Build Performance

from transforms.api import transform_df, Input, Output, configure, incremental

# Use incremental for append-only data — processes only new rows
@incremental()
@transform_df(
    Output("/Company/datasets/events_processed"),
    events=Input("/Company/datasets/raw_events"),
)
def process_events(events):
    return events.filter(events.event_type.isNotNull())

# Tune Spark resources for heavy aggregations
@configure(profile=["DRIVER_MEMORY_LARGE", "EXECUTOR_MEMORY_LARGE"])
@transform_df(
    Output("/Company/datasets/daily_summary"),
    data=Input("/Company/datasets/large_table"),
)
def daily_summary(data):
    from pyspark.sql import functions as F
    return data.groupBy("date", "region").agg(
        F.sum("revenue").alias("total_revenue"),
        F.countDistinct("user_id").alias("unique_users"),
    )

Step 5: Connection Pooling

import requests
from requests.adapters import HTTPAdapter
from urllib3.util.retry import Retry

session = requests.Session()
adapter = HTTPAdapter(
    pool_connections=10,
    pool_maxsize=20,
    max_retries=Retry(total=3, backoff_factor=0.5, status_forcelist=[429, 500, 502, 503]),
)
session.mount("https://", adapter)

Output

• Maximum page size pagination reducing API call count
• TTL-based caching for repeated object reads
• Batch search replacing individual GET calls
• Optimized Spark transforms with @configure and @incremental

Error Handling

Resources

• Transforms @configure
• Incremental Transforms
• cachetools

Next Steps

For cost optimization, see palantir-cost-tuning.