--- title: "How to read a JSONL file in Python (faster than pandas)" date: "2026-06-08" author: "Al Brown" category: "Engineering" excerpt: "Read a JSONL file into a DataFrame with chDB, a drop-in replacement for pandas. Change one import line and your existing pandas code runs on ClickHouse's engine, so it stays fast as files grow." --- # How to read a JSONL file in Python (faster than pandas) To read a JSONL file in Python, use [chDB](/resources/engineering/what-is-chdb) as a drop-in replacement for pandas. Change one import line and the code you already write keeps working: ```python import chdb.datastore as pd df = pd.read_json("data/orders.jsonl", lines=True) ``` `df` supports the pandas API you already use (indexing, filtering, `groupby`, `merge`), but the work runs on ClickHouse's engine instead of pandas, so it stays fast as files grow. There is no server to start and no separate load step. ```bash pip install chdb ``` Prefer the command line? See [how to run SQL on a JSONL file](/resources/engineering/run-sql-on-jsonl-file) to do this with clickhouse local. ## Read a JSONL file into a DataFrame `pd.read_json` with `lines=True` reads json-lines files the way you expect. Point it at a path and print the result: ```python import chdb.datastore as pd df = pd.read_json("data/orders.jsonl", lines=True) print(df) print(df.dtypes) ``` ```text order_id country status amount customer skus 0 1 GB paid 49.99 [101, gold] [sku-a, sku-b] 1 2 AU cart 0.00 [102, silver] [sku-c] 2 3 GB paid 19.50 [103, gold] [sku-a] 3 4 IN paid 5.00 [104, bronze] [sku-d, sku-e, sku-f] 4 5 AU paid 99.00 [105, gold] [] order_id int64 country str status str amount float64 customer object skus object dtype: object ``` Look at the dtypes. chDB inferred them from the data: `order_id` is an integer, `amount` is a float. Nested fields like `customer` and array fields like `skus` come through as object columns, ready to work with in pandas. One thing is different from pandas under the hood: `df` is not the whole file loaded into memory. It is a lazy, ClickHouse-backed object that records what you asked for and runs it on ClickHouse's engine when you materialize a result, whether you print it, take `len()`, or call `.to_pandas()`. That is why it stays fast on large files. ## Filter and aggregate the way you already do The pandas you write does not change. Filter with a boolean mask, group, and aggregate: ```python import chdb.datastore as pd df = pd.read_json("data/orders.jsonl", lines=True) revenue = (df[df["status"] == "paid"] .groupby("country")["amount"].sum() .sort_values(ascending=False)) print(revenue) ``` ```text country AU 99.00 GB 69.49 IN 5.00 Name: amount, dtype: float64 ``` Same syntax as pandas, same result. chDB compiles the whole chain into one optimized query and only scans the columns the aggregation touches, rather than materializing the full file first. ## Access nested fields and explode arrays When you need to work with the nested `customer` struct or the `skus` array, materialize the small file with `.to_pandas()` first and then use standard pandas: ```python pdf = df.to_pandas() # customer comes back as [id, tier]; index 1 is the tier pdf["tier"] = pdf["customer"].apply(lambda c: c[1]) print(pdf[["order_id", "country", "status", "tier"]].to_string(index=False)) ``` ```text order_id country status tier 1 GB paid gold 2 AU cart silver 3 GB paid gold 4 IN paid bronze 5 AU paid gold ``` The `skus` column is an array per row. Use `.explode()` to flatten it into one row per SKU: ```python paid = pdf[pdf["status"] == "paid"][["country", "skus"]].explode("skus") print(paid.dropna(subset=["skus"]).reset_index(drop=True)) ``` ```text country skus 0 GB sku-a 1 GB sku-b 2 GB sku-a 3 IN sku-d 4 IN sku-e 5 IN sku-f ``` The pattern is: do the heavy filter and aggregation on the chDB object (fast, ClickHouse-backed), then `.to_pandas()` the result and continue in ordinary pandas for struct unpacking or explode operations. ## Hand off to real pandas when you need it When a library downstream needs an actual pandas DataFrame (scikit-learn, a plotting call, anything that mutates in place), call `.to_pandas()` to materialize one: ```python df = pd.read_json("data/orders.jsonl", lines=True) pdf = df.to_pandas() # a real pandas.DataFrame, in memory ``` The usual pattern is to do the heavy filtering and aggregation on the chDB object, where it is fast, then `.to_pandas()` the small result and continue in ordinary pandas. ## Is it faster than `pandas.read_json`? On a file big enough that scanning is the actual cost, yes. `pandas.read_json(lines=True)` decodes every line into Python objects and builds a DataFrame before it can group anything, while chDB streams the parse and aggregation in compiled, multi-threaded code and only reads what you ask for. The aggregation above on a 2M-row JSONL file, best-of-3 with a warm page cache on an Apple M4 Pro (14 cores, 24 GB RAM, macOS); chDB 4.1.8, Python 3.14; best-of-3, warm: ```text import pandas as pd: 1.555s import chdb.datastore as pd: 0.123s speedup: 12.6x ``` About 12x here, from the same code with one import changed. The back-to-back ratio is the robust claim; absolute times shift with cache state and concurrent load. The gap widens as files grow, because plain pandas decodes and holds the full file in memory before it can group anything, while chDB streams and aggregates in one pass. The small case can cut the other way: on a tiny JSONL (a few thousand rows) plain pandas is as fast or faster, because chDB pays a fixed per-query startup cost that dominates when there is barely any data. Reach for chDB when the file is large, or when you want one import to cover both. ## Works in Jupyter Printing a chDB object renders it as a table in a notebook, and `.to_pandas()` feeds straight into `.plot()`, joins, or any pandas code. The companion folder ships a `run.ipynb` you can open and run cell by cell. ## Run it yourself The complete, runnable example is here, with `generate.sh` to create the sample JSONL files plus `run.ipynb` and a `run.py` mirror containing the exact code above: [github.com/ClickHouse/examples/tree/main/local-analytics/chdb-jsonl](https://github.com/ClickHouse/examples/tree/main/local-analytics/chdb-jsonl) Working with other shapes of JSON? See [read a JSON file in Python](/resources/engineering/read-json-file-python) and [read an NDJSON file in Python](/resources/engineering/read-ndjson-file-python). They use the same `import chdb.datastore as pd` pattern. If you want to avoid re-parsing JSONL on every read, [convert it to Parquet](/resources/engineering/convert-jsonl-to-parquet) once and every later read gets faster. The same code scales from a JSONL file on your laptop to a server to [ClickHouse Cloud](https://clickhouse.com/cloud) with no rewrite, so there is nothing to re-learn when the data outgrows one machine.