scripts.md

type	Reference
title	e2sar-daos scripts
description	The bash and Python scripts that drive the apps on the JLab test nodes: the e2e bench, its rate sweep, the layout sweep and the site receiver.
tags	scripts e2sar_e2e_bench e2sar_site2ebpf_recv
sources	id resource title origin https://github.com/JeffersonLab/e2sar-daos/blob/0d85d00/scripts/README.md scripts/README.md at 0d85d00

e2sar-daos scripts

Bash scripts that drive the applications for tests on the JLab ebpf nodes. Each one prints its usage when run without arguments, and the comment at its top lists its settings.

Script	Where it runs	What it does
e2sar_e2e_bench.sh	Receiver and sender node, one role each	One end-to-end run: e2sar_evio_send (the frames of an EVIO file, -F, one per event unless SEND_ARGS="-f F") → [LB] → e2sar_daos_recv → DAOS. Throughput, latency per stage, and a digest check of every event. See e2e-bench
e2sar_e2e_sweep.sh	A machine that can SSH to both nodes	A rate sweep: runs e2sar_e2e_bench.sh on both nodes for each rate and repetition. See 2026-10-20-e2e-n2n-sweep section 7
e2sar_layout_sweep.py	A machine that can SSH to both nodes	The tuning sweep of e2sar_daos_recv (reassembly and dequeue threads, in-flight operations, batch flush) for the scenarios of 2026-10-29-receiver-tuning, one JSON line per run
e2sar_site2ebpf_recv.sh	ebpf2201 and ebpf2203	Runs one node as a receiver for a remote sender (Perlmutter, or another site; SITE only picks the log directory): pre-flight checks, shared run directory, e2sar_daos_recv (with -v) into a new container <CONT_LABEL>_<UTC epoch s> per run, drop checks after Ctrl-C, then verify N, runs (the runs in DAOS, from the run catalog) and clean [RUNID] (destroy one run's container, or every <CONT_LABEL>_<epoch s> container). See perlmutter
e2sar_live_pipeline.sh	A machine that can SSH to every node (the laptop)	The live pipeline of streaming-workflow for a remote sender through an LB. start: a new container LABEL_<UTC epoch s>, on each receiver node e2sar_nicmon.sh then e2sar_daos_recv -c, on the reader node the DAOS monitor and its page, e2sar_daos_read --live $READ_ARGS (--min-free 50: the reader pauses below 50 GiB free, see e2sar_daos_read) and the EVIO viewer. stop: receivers (they drain), reader, monitor, pages, NIC monitors; data kept, logs to LOGDIR/<run>/. status: processes, LB senders and workers, events, reader files, missing, NIC report, free space, and READER PAUSED when the reader is waiting for space. recycle: if the run is at least IDLE_S (3600 s) old and has no event at all, delete it (container, empty folder, monitor file; logs kept in LOGDIR/empty_<run>/) and start a new one; a run with any event is never deleted. watch: status and recycle every EVERY_S (3600 s), warning below MIN_FREE_GB (150). Never frees the LB. See live-monitoring
e2sar_nicmon.sh	Every receiver node, for every run	Samples the NIC (ethtool -S: packets at the port and to the host, discards, out of buffer, steer-missed), IP, UDP and TCP counters (/proc/net/snmp, read by name) and softnet drops once a second: start DIR (snapshot, then a background sampler into DIR/nic_<host>.log), stop DIR (stop, snapshot), report DIR (each drop counter's change, the seconds in which one moved, and the seconds in which more than GAP packets (3000) reached the port but not a UDP socket). e2sar_e2e_bench.sh and e2sar_site2ebpf_recv.sh run it around the receiver. No root needed. On the ebpf nodes the data port also carries NFS ($HOME) and SSH, so check Tcp.InSegs before calling a gap a loss
evio_hit_compression.py	Anywhere with numpy and pyarrow (evio-interactive-view's .venv)	Reads the first N frames (20,000) of an EVIO v6 file or --live bundle of fADC250 time frames and prints how far they compress, and how fast on one thread: lz4 and zstd on the raw bytes and per frame, Parquet (three layouts), and hit-aware columns (time step within a payload bank, channel, charge, count). Also the entropy of each hit-word field. --tmp DIR for the Parquet test files (removed after). See hit-compression
utc.py	Anywhere with Python ≥ 3.6 (standard library only)	Turns UTC epoch times into dates: seconds (1793817098 --> UTC 2026-11-04 18:31:38), log timestamps with a fraction, ms/µs/ns by digit count (e.g. the LB's predictedeventnum), the seconds in run and bundle names, and DAOS object ids hi.lo (kind, slot, second; see daos-layout). With no arguments it reads lines from stdin and adds the date after each time, e.g. tail recv.log | scripts/utc.py. -h for usage, --local adds local time

Before running them

• Build first. The scripts run the apps from the build next to them, ../build/bin/ (meson compile -C build). Set E2SAR_DAOS_BUILD to use another build. Run the copy of a script in the checkout whose build you want, e.g. in a separate worktree.
• Environment. They read ~/.bashrc for E2SAR_BIN (the E2SAR tools, such as lbadm) and, through the LB, the instance URI (EJFAT_URI). e2sar_site2ebpf_recv.sh takes the URI as EJFAT_URI (passed in or exported) and its listen address from MY_IP. See the E2SAR environment and, for the LB, lb-test step 1.
• Settings. Every setting is an environment variable with a default for the ebpf testbed, listed at the top of the script, e.g. DEQ=16 scripts/e2sar_site2ebpf_recv.sh start.
• Shared $HOME. The two ebpf nodes can't SSH to each other. The roles meet through files in $RUN, under the NFS $HOME both nodes share.
• bash. The ebpf login shell is tcsh: start bash first, or run a script as bash scripts/<name>.sh.

Examples

# end-to-end bench through the LB: receiver first, then the sender, with the same TAG
bash ~/daq/e2sar-daos/scripts/e2sar_e2e_bench.sh recv run1 -m lb      # on ebpf2203
bash ~/daq/e2sar-daos/scripts/e2sar_e2e_bench.sh send run1 -m lb      # on ebpf2201

# a node-to-node rate sweep, 3 repetitions per rate, from a laptop
scripts/e2sar_e2e_sweep.sh 1020a "20 25 30 35 40 45 50" 3

# the Perlmutter receivers: ebpf2201 first, then ebpf2203
~/daq/e2sar-daos/scripts/e2sar_site2ebpf_recv.sh start

The analysis step of the bench, bin/e2sar_e2e_analyze.py, stays in bin/ next to the apps, and e2sar_e2e_bench.sh calls it from there.