# ========= Copyright 2026 @ Strukto.AI All Rights Reserved. ========= # Licensed under the Apache License, Version 2.0 (the "License"); # you may not use this file except in compliance with the License. # You may obtain a copy of the License at # # http://www.apache.org/licenses/LICENSE-2.0 # # Unless required by applicable law or agreed to in writing, software # distributed under the License is distributed on an "AS IS" BASIS, # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. # See the License for the specific language governing permissions and # limitations under the License. # ========= Copyright 2026 @ Strukto.AI All Rights Reserved. ========= import asyncio import os import time import uuid from dotenv import load_dotenv from mirage import MountMode, Workspace from mirage.resource.hf_buckets import HfBucketsConfig, HfBucketsResource load_dotenv(".env.development") config = HfBucketsConfig( bucket=os.environ["HF_BUCKET_NAME"], token=os.environ["HF_TOKEN"], ) resource = HfBucketsResource(config) ws = Workspace({"/hf/": resource}, mode=MountMode.WRITE) def ops_summary() -> str: records = ws.ops.records total = sum(r.bytes for r in records) return f"{len(records)} ops, {total} bytes transferred" def show_plan(label: str, dr) -> None: print(f"\n--- plan: {label} ---") print(f" network_read: {dr.network_read} cache_read: {dr.cache_read}") print(f" read_ops: {dr.read_ops} cache_hits: {dr.cache_hits} " f"precision: {dr.precision}") async def main(): print("=== not-found errors show the full virtual path ===") for cmd in ("cat /hf/__nf_missing__.txt", "head /hf/__nf_missing__.txt", "stat /hf/__nf_missing__.txt"): result = await ws.execute(cmd) print(f"$ {cmd}") print(f" exit={result.exit_code} " f"{(await result.stderr_str()).strip()}") # ── discover structure ──────────────────────────── print("=== ls /hf/ ===") r = await ws.execute("ls /hf/") print(await r.stdout_str()) # ── tree ────────────────────────────────────────── print("=== tree /hf/ ===") r = await ws.execute("tree /hf/") print(await r.stdout_str()) # ── stat (root + file) ────────────────────────────── print("=== stat /hf (root) ===") r = await ws.execute("stat /hf") print(f" {(await r.stdout_str()).strip()}") print("\n=== stat /hf/example.json ===") r = await ws.execute("stat /hf/example.json") print(f" {(await r.stdout_str()).strip()}") # ── cat json ────────────────────────────────────── print("\n=== cat /hf/example.json | head -n 10 ===") r = await ws.execute("cat /hf/example.json | head -n 10") print(await r.stdout_str()) # ── head / tail on jsonl ────────────────────────── print("=== head -n 3 /hf/example.jsonl ===") r = await ws.execute("head -n 3 /hf/example.jsonl") print((await r.stdout_str())[:300]) print("\n=== tail -n 2 /hf/example.jsonl ===") r = await ws.execute("tail -n 2 /hf/example.jsonl") print((await r.stdout_str())[:300]) # ── wc ──────────────────────────────────────────── print("\n=== wc -l /hf/example.jsonl ===") r = await ws.execute("wc -l /hf/example.jsonl") print(f" {(await r.stdout_str()).strip()}") # ── grep ────────────────────────────────────────── print("\n=== grep -c mirage /hf/example.jsonl ===") r = await ws.execute("grep -c mirage /hf/example.jsonl") print(f" count: {(await r.stdout_str()).strip()}") print("\n=== grep mirage /hf/example.jsonl | head -n 3 ===") r = await ws.execute("grep mirage /hf/example.jsonl | head -n 3") lines = (await r.stdout_str()).strip().splitlines() for ln in lines: print(f" {ln[:100]}...") # ── find ────────────────────────────────────────── print("\n=== find /hf/ -name '*.json' ===") r = await ws.execute("find /hf/ -name '*.json'") print(await r.stdout_str()) print("=== find /hf/ -name '*.parquet' ===") r = await ws.execute("find /hf/ -name '*.parquet'") print(await r.stdout_str()) # ── jq ──────────────────────────────────────────── print("=== jq .metadata /hf/example.json ===") r = await ws.execute("jq .metadata /hf/example.json") print(f" {(await r.stdout_str()).strip()[:200]}") print("\n=== jq -r '.departments[].teams[].name'" " /hf/example.json ===") r = await ws.execute('jq ".departments[].teams[].name" /hf/example.json') print(f" {(await r.stdout_str()).strip()}") # ── pipelines ───────────────────────────────────── print("\n=== cat example.jsonl | grep queue-operation" " | sort | uniq | wc -l ===") r = await ws.execute("cat /hf/example.jsonl" " | grep queue-operation | sort | uniq | wc -l") print(f" unique lines: {(await r.stdout_str()).strip()}") # ── cd + relative paths ─────────────────────────── print("\n=== pwd ===") r = await ws.execute("pwd") print(f" {(await r.stdout_str()).strip()}") print('\n=== cd /hf ===') r = await ws.execute("cd /hf") print(f" exit={r.exit_code}") print("\n=== pwd (after cd) ===") r = await ws.execute("pwd") print(f" {(await r.stdout_str()).strip()}") print("\n=== ls (relative) ===") r = await ws.execute("ls") print(await r.stdout_str()) print("=== head -n 3 example.json (relative) ===") r = await ws.execute("head -n 3 example.json") print(await r.stdout_str()) # ── streaming & barrier scenarios ───────────────── print("\n=== cat | grep | head (streaming drain) ===") r = await ws.execute("cat /hf/example.jsonl | grep queue | head -n 3") lines = (await r.stdout_str()).strip().splitlines() print(f" got {len(lines)} lines (expected 3)") print("\n=== grep -q && echo (barrier VALUE) ===") r = await ws.execute('grep -q queue /hf/example.jsonl && echo "found"') print(f" stdout: {(await r.stdout_str()).strip()}") print(f" exit: {r.exit_code}") print("\n=== grep -q || echo (barrier OR) ===") r = await ws.execute('grep -q NONEXISTENT_STRING /hf/example.jsonl' ' || echo "not found"') print(f" stdout: {(await r.stdout_str()).strip()}") print(f" exit: {r.exit_code}") print("\n=== grep ; grep (semicolon materialization) ===") r = await ws.execute("grep -c queue /hf/example.jsonl" "; grep -c mirage /hf/example.jsonl") print(f" stdout: {(await r.stdout_str()).strip()}") print("\n=== grep missing ; echo $? (semicolon exit code) ===") r = await ws.execute("grep NONEXISTENT_STRING /hf/example.jsonl" "; echo $?") print(f" stdout: {(await r.stdout_str()).strip()}") print("\n=== cat nonexistent 2>&1 | head (stderr in pipe) ===") r = await ws.execute("cat /hf/nonexistent_file 2>&1 | head -n 1") print(f" stdout: {(await r.stdout_str()).strip()}") print(f" exit: {r.exit_code}") print("\n=== cat nonexistent | cat (no merge: error in stderr) ===") r = await ws.execute("cat /hf/nonexistent_file | cat") print(f" stdout: '{(await r.stdout_str()).strip()}' (expect empty)") print(f" stderr: '{(await r.stderr_str()).strip()[:80]}'") print("\n=== cat nonexistent 2>&1 | cat (no double-emit) ===") r = await ws.execute("cat /hf/nonexistent_file 2>&1 | cat") out = (await r.stdout_str()).strip() err = (await r.stderr_str()).strip() print(f" stdout: '{out[:80]}'") print(f" stderr: '{err}' (expect empty: no double-emit)") print("\n=== cat large 2>&1 | wc -l (streams real payload) ===") r = await ws.execute("wc -l /hf/example.jsonl") expected = int((await r.stdout_str()).strip().split()[0]) r = await ws.execute("cat /hf/example.jsonl 2>&1 | wc -l") got = int((await r.stdout_str()).strip()) print(f" expected: {expected} got: {got} " f"{'OK' if got == expected else 'MISMATCH'}") print("\n=== cat | sort | uniq | wc -l (full pipeline) ===") r = await ws.execute("cat /hf/example.jsonl | sort | uniq | wc -l") print(f" unique lines: {(await r.stdout_str()).strip()}") # ── background job scenarios ──────────────────── print("\n=== grep -c & echo kicked off; wait (bg job) ===") r = await ws.execute("grep -c queue /hf/example.jsonl &" " echo 'kicked off'; wait") print(f" stdout: {(await r.stdout_str()).strip()}") print("\n=== sleep 0 & cat (bg doesn't consume stdin) ===") r = await ws.execute("sleep 0 & cat /hf/example.json | head -n 1") print(f" stdout: {(await r.stdout_str()).strip()}") print("\n=== cat nonexistent & echo ok (bg error handled) ===") r = await ws.execute("cat /hf/nonexistent_file & echo ok; wait; echo done") print(f" stdout: {(await r.stdout_str()).strip()}") print(f" exit: {r.exit_code}") print("\n=== multiple bg: grep & wc & wait (parallel) ===") r = await ws.execute("grep -c queue /hf/example.jsonl &" " wc -l /hf/example.jsonl &" " wait; echo all done") print(f" stdout: {(await r.stdout_str()).strip()}") # ── lazy stdin in loops ─────────────────────────── print("\n=== head -n 5 | while read; do echo (bounded loop) ===") r = await ws.execute("cat /hf/example.jsonl | head -n 5" " | while read LINE; do echo got; done | wc -l") print(f" iterations: {(await r.stdout_str()).strip()} (expected 5)") print("\n=== while read; break (early exit) ===") r = await ws.execute("cat /hf/example.jsonl | head -n 100" " | while read LINE; do echo first; break; done") out = (await r.stdout_str()).strip().splitlines() print(f" stdout lines: {len(out)} (expected 1) exit={r.exit_code}") print("\n=== for x in a b c; do read LINE (loop reads buffer) ===") r = await ws.execute( "cat /hf/example.jsonl | head -n 3" " | for x in a b c; do read LINE; echo \"$x:${LINE:0:30}\"; done") for line in (await r.stdout_str()).strip().splitlines(): print(f" {line}") # ── quoting / escaping ──────────────────────────── print("\n=== echo \"\\$X\" (escaped dollar stays literal) ===") await ws.execute("export X=expanded") r = await ws.execute('echo "\\$X"') print(f" stdout: {(await r.stdout_str()).strip()!r} (expect '$X')") print("\n=== echo \"$X\" (unescaped dollar expands) ===") r = await ws.execute('echo "$X"') print(f" stdout: {(await r.stdout_str()).strip()!r} (expect 'expanded')") print("\n=== echo '$X' (single quotes keep $X literal) ===") r = await ws.execute("echo '$X'") print(f" stdout: {(await r.stdout_str()).strip()!r} (expect '$X')") print("\n=== cat \"$DIR/example.json\" (env var in path) ===") await ws.execute("export DIR=/hf") r = await ws.execute('cat "$DIR/example.json" | head -n 3') out = (await r.stdout_str()).strip().splitlines() print(f" first lines: {out}") print("\n=== cat $(echo /hf/example.json) | head -n 1" " (command sub as path) ===") r = await ws.execute("cat $(echo /hf/example.json) | head -n 1") print(f" stdout: {(await r.stdout_str()).strip()}") print("\n=== grep \"$(echo queue)\" /hf/example.jsonl" " | wc -l (sub as pattern) ===") r = await ws.execute('grep "$(echo queue)" /hf/example.jsonl | wc -l') print(f" count: {(await r.stdout_str()).strip()}") # ── PROVISION: estimate cost without executing ──────────────────── # Returns a ProvisionResult with network_read / cache_read / precision # instead of running the command. Provisioned commands: # file_read_provision: cat, wc (full-file read, EXACT) # head_tail_provision: head, tail (EXACT for -c, RANGE for -n) # metadata_provision : find, ls, stat (0 bytes, EXACT) # grep / jq : inline provisioners in their command files # When a file is already in the workspace cache, the workspace # automatically moves network_read into cache_read and zeros # network_read. That's why we clear the cache first to see the # "first-time" estimate, then re-plan after caching to see the diff. print("\n=== PROVISION (plan without executing) ===") await ws.cache.clear() before_plan = ops_summary() # ── single-command plans (cache cleared, so network_read shows up) ── dr = await ws.execute("cat /hf/example.json", provision=True) show_plan("cat /hf/example.json (file_read_provision)", dr) dr = await ws.execute("wc -l /hf/example.jsonl", provision=True) show_plan("wc -l /hf/example.jsonl (file_read_provision)", dr) dr = await ws.execute("head -c 100 /hf/example.jsonl", provision=True) show_plan("head -c 100 /hf/example.jsonl (byte budget, EXACT)", dr) dr = await ws.execute("head -n 5 /hf/example.jsonl", provision=True) show_plan("head -n 5 /hf/example.jsonl (line budget, RANGE)", dr) print(f" range: [{dr.network_read_low}, {dr.network_read_high}]") dr = await ws.execute("ls /hf/", provision=True) show_plan("ls /hf/ (metadata_provision, 0 bytes)", dr) dr = await ws.execute("stat /hf/example.json", provision=True) show_plan("stat /hf/example.json (metadata_provision)", dr) dr = await ws.execute("grep mirage /hf/example.jsonl", provision=True) show_plan("grep mirage /hf/example.jsonl (grep_provision)", dr) dr = await ws.execute("jq .metadata /hf/example.json", provision=True) show_plan("jq .metadata /hf/example.json (jq_provision)", dr) # ── compound plans ── dr = await ws.execute("cat /hf/example.jsonl | head -n 3", provision=True) print("\n--- plan: cat | head -n 3 (pipeline with children) ---") print(f" op: {dr.op} children: {len(dr.children)} " f"precision: {dr.precision}") print(f" network_read: {dr.network_read} cache_read: {dr.cache_read}") for c in dr.children: print(f" {c.command}: net={c.network_read} " f"cache={c.cache_read} {c.precision}") dr = await ws.execute("grep mirage /hf/example.jsonl && echo found", provision=True) print("\n--- plan: grep ... && echo found (compound) ---") print(f" op: {dr.op} network_read: {dr.network_read}") for c in dr.children: print(f" {c.command}: net={c.network_read} {c.precision}") print(f"\n Stats before plans: {before_plan}") print(f" Stats after plans: {ops_summary()} (planning is read-free)") # ── cache-aware plan: same command after caching the file ── print("\n--- caching: cat /hf/example.json | wc -c ---") r = await ws.execute("cat /hf/example.json | wc -c") print(f" bytes: {(await r.stdout_str()).strip()}") dr = await ws.execute("cat /hf/example.json", provision=True) print("\n--- plan after cache: cat /hf/example.json ---") print(f" network_read: {dr.network_read} " f"cache_read: {dr.cache_read} cache_hits: {dr.cache_hits}") print(" (network shifted to cache: file is hot)") # ── chunk-level streaming + multi-stage pipe backpressure ───────── print("\n=== STREAMING (single command) ===") target = "/hf/example.jsonl" r = await ws.execute(f"stat -c '%s' {target}") size = int((await r.stdout_str()).strip()) print(f" object size: {size:,} bytes") async def measure(label: str, cmd: str) -> None: before = sum(rec.bytes for rec in ws.ops.records) t0 = time.monotonic() r = await ws.execute(cmd) dt = time.monotonic() - t0 net = sum(rec.bytes for rec in ws.ops.records) - before head = (await r.stdout_str()).strip().splitlines() first = head[0][:48] if head else "" print(f" {label:42s} bytes={net:>10,} t={dt:4.2f}s " f"lines={len(head):>4} out0={first!r}") await ws.cache.clear() await measure("head -n 1 (line-streamed)", f"head -n 1 {target}") await ws.cache.clear() await measure("head -c 100 (byte-range)", f"head -c 100 {target}") await ws.cache.clear() await measure("grep -m 1 (early-exit)", f"grep -m 1 mirage {target}") print("\n=== STREAMING CHAIN (multi-stage pipe backpressure) ===") await ws.cache.clear() await measure("cat | head -n 1", f"cat {target} | head -n 1") await ws.cache.clear() await measure("cat | tr A-Z a-z | head -n 1", f"cat {target} | tr A-Z a-z | head -n 1") await ws.cache.clear() await measure("cat | grep mirage | head -n 1", f"cat {target} | grep mirage | head -n 1") await ws.cache.clear() await measure("4-stage: cat|tr|grep|head -n 1", f"cat {target} | tr A-Z a-z | grep mirage | head -n 1") await ws.cache.clear() await measure( "5-stage: cat|tr|grep|head|wc -l", f"cat {target} | tr A-Z a-z | grep mirage | head -n 1 " "| wc -l") await ws.cache.clear() await measure("non-cancellable: cat | wc -l", f"cat {target} | wc -l") # WRITE + REMOVE flow. HF Buckets silently drops zero-byte uploads, # so we use ws.ops.write to push non-empty bytes (touch would no-op). # Demonstrates parent-dir index cache invalidation: without it, `ls` # after a write would return stale entries. test_file = f"/hf/test-{uuid.uuid4().hex[:8]}.txt" test_file.rsplit("/", 1)[-1] print(f"\n=== WRITE + REMOVE FLOW (using {test_file}) ===") print(f"\n--- write '{test_file}' (14 bytes) ---") await ws.ops.write(test_file, b"Hello, Mirage!") print(" written") print(f"\n--- stat {test_file} (should succeed) ---") r = await ws.execute(f"stat {test_file}") print(f" {(await r.stdout_str()).strip()}") print(f"\n--- cat {test_file} ---") r = await ws.execute(f"cat {test_file}") print(f" {(await r.stdout_str()).strip()!r}") print(f"\n--- rm {test_file} ---") r = await ws.execute(f"rm {test_file}") print(f" exit: {r.exit_code}") print(f"\n--- stat {test_file} (should fail: not found) ---") r = await ws.execute(f"stat {test_file}") err = (await r.stderr_str()).strip()[:80] print(f" exit: {r.exit_code} stderr: {err}") # touch on an existing file is still tested (no-op when c=False but # file exists, since exists() short-circuits). print("\n--- touch on existing file (no-op) ---") r = await ws.execute("touch /hf/example.json") print(f" exit: {r.exit_code}") print(f"\nStats: {ops_summary()}") if __name__ == "__main__": asyncio.run(main())