jsharkey13 · January 18, 2024 11:37 · jsharkey13 · Jan 18, 2024
diff --git a/zpool_iostat_latency_graphs.py b/zpool_iostat_latency_graphs.py
 # zpool disk latency visualisation
 #
 # This script parses the output of the zpool stats command:
 #      zpool iostat -pvwH
 # which provides per-disk latency information for reads and writes.
 #
 # Run the command:
 #
 #     zpool iostat -pvwH > output.tsv
 #
 # and copy the resulting file to the same folder as this script, then run
 # this script.
 #
 # The ["tank"] list can be replaced by any names to exclude (i.e. the pool and groups)
 # and "zpool" can be changed to an appropriate title.
 #


 import matplotlib.pyplot as plt
 import numpy as np

 __all__ = ["parse_output", "display_graphs"]

 COLUMN_HEADINGS = ["total_wait:read", "total_wait:write",
                   "disk_wait:read", "disk_wait:write",
                   "syncq_wait:read", "syncq_wait:write",
                   "asyncq_wait:read", "asyncq_wait:write",
                   "scrub", "trim"]

 BUCKETS = ["1ns", "3ns", "7ns", "15ns", "31ns", "63ns", "127ns", "255ns", "511ns",
           "1us", "2us", "4us", "8us", "16us", "32us", "65us", "131us", "262us", "524us",
           "1ms", "2ms", "4ms", "8ms", "16ms", "33ms", "67ms", "134ms", "268ms", "536ms",
           "1s", "2s", "4s", "8s", "17s", "34s", "68s", "137s"]

 COLOURS = ["red", "darkorange", "gold", "greenyellow", "seagreen", "turquoise", "royalblue", "rebeccapurple"]


 def _parse_histogram_rows(tsv_rows):
    results = dict()
    for row in tsv_rows:
        bucket, *data = row.split("\t")
        assert len(data) == len(COLUMN_HEADINGS), f"Expected {len(COLUMN_HEADINGS)} columns, found {len(data)}!"

        results[bucket] = dict(zip(COLUMN_HEADINGS, map(int, data)))

    assert len(results) == len(BUCKETS), f"Expected {len(BUCKETS)} rows, found {len(results)}!"
    return results


 def parse_output(tsv_value):
    """Parse the output of `zpool iostat -pwvH` into a dictionary."""

    results = dict()
    for section_value_str in tsv_value.split("\n\n"):
        section_value = section_value_str.strip()
        if section_value == "":
            continue

        title, *histogram_rows = section_value.split("\n")
        results[title] = _parse_histogram_rows(histogram_rows)

    return results


 def display_graphs(zpool_stats, *, name="", include_disks=None):
    """Plot the latency stats for a group of disks."""

    if include_disks is None:
        include_disks = zpool_stats.keys()

    for col in COLUMN_HEADINGS:
        n_disks = len(include_disks)
        width = round(1 / (n_disks + 2), 3)
        bin_positions = np.arange(len(BUCKETS))

        fig, ax = plt.subplots()
        fig.set_size_inches(20, 8)
        figtitle = f"{name} {col}"
        fig.suptitle(figtitle, fontsize=20)

        handles = []
        for i, disk in enumerate(include_disks):

            data = [r[col] for r in zpool_stats[disk].values()]
            if sum(data) == 0:
                # Skip all-blank groups.
                continue

            x_positions = bin_positions + (i-(n_disks // 2))*width  # Shift bars around centre of the bin.

            h = ax.bar(x_positions, data, width, color=COLOURS[i])
            handles.append(h)

            ax.set_xticks(range(len(BUCKETS))[::2])  # Too crowded with every bin labelled.
            ax.set_xticklabels(BUCKETS[::2])
            ax.semilogy()

        ax.legend(handles=handles, labels=include_disks)

        plt.show()


 if __name__ == "__main__":
    # Run `zpool iostat -pwvH > output.tsv`.
    with open("output.tsv") as f:
        stats = parse_output(f.read())
        disks = [d for d in stats.keys() if d not in ["tank"]]  # Exclude whole-pool stats?
        display_graphs(stats, name="zpool", include_disks=disks)
	# zpool disk latency visualisation
	#
	# This script parses the output of the zpool stats command:
	# zpool iostat -pvwH
	# which provides per-disk latency information for reads and writes.
	#
	# Run the command:
	#
	# zpool iostat -pvwH > output.tsv
	#
	# and copy the resulting file to the same folder as this script, then run
	# this script.
	#
	# The ["tank"] list can be replaced by any names to exclude (i.e. the pool and groups)
	# and "zpool" can be changed to an appropriate title.
	#


	import matplotlib.pyplot as plt
	import numpy as np

	__all__ = ["parse_output", "display_graphs"]

	COLUMN_HEADINGS = ["total_wait:read", "total_wait:write",
	"disk_wait:read", "disk_wait:write",
	"syncq_wait:read", "syncq_wait:write",
	"asyncq_wait:read", "asyncq_wait:write",
	"scrub", "trim"]

	BUCKETS = ["1ns", "3ns", "7ns", "15ns", "31ns", "63ns", "127ns", "255ns", "511ns",
	"1us", "2us", "4us", "8us", "16us", "32us", "65us", "131us", "262us", "524us",
	"1ms", "2ms", "4ms", "8ms", "16ms", "33ms", "67ms", "134ms", "268ms", "536ms",
	"1s", "2s", "4s", "8s", "17s", "34s", "68s", "137s"]

	COLOURS = ["red", "darkorange", "gold", "greenyellow", "seagreen", "turquoise", "royalblue", "rebeccapurple"]


	def _parse_histogram_rows(tsv_rows):
	results = dict()
	for row in tsv_rows:
	bucket, *data = row.split("\t")
	assert len(data) == len(COLUMN_HEADINGS), f"Expected {len(COLUMN_HEADINGS)} columns, found {len(data)}!"

	results[bucket] = dict(zip(COLUMN_HEADINGS, map(int, data)))

	assert len(results) == len(BUCKETS), f"Expected {len(BUCKETS)} rows, found {len(results)}!"
	return results


	def parse_output(tsv_value):
	"""Parse the output of `zpool iostat -pwvH` into a dictionary."""

	results = dict()
	for section_value_str in tsv_value.split("\n\n"):
	section_value = section_value_str.strip()
	if section_value == "":
	continue

	title, *histogram_rows = section_value.split("\n")
	results[title] = _parse_histogram_rows(histogram_rows)

	return results


	def display_graphs(zpool_stats, *, name="", include_disks=None):
	"""Plot the latency stats for a group of disks."""

	if include_disks is None:
	include_disks = zpool_stats.keys()

	for col in COLUMN_HEADINGS:
	n_disks = len(include_disks)
	width = round(1 / (n_disks + 2), 3)
	bin_positions = np.arange(len(BUCKETS))

	fig, ax = plt.subplots()
	fig.set_size_inches(20, 8)
	figtitle = f"{name} {col}"
	fig.suptitle(figtitle, fontsize=20)

	handles = []
	for i, disk in enumerate(include_disks):

	data = [r[col] for r in zpool_stats[disk].values()]
	if sum(data) == 0:
	# Skip all-blank groups.
	continue

	x_positions = bin_positions + (i-(n_disks // 2))*width # Shift bars around centre of the bin.

	h = ax.bar(x_positions, data, width, color=COLOURS[i])
	handles.append(h)

	ax.set_xticks(range(len(BUCKETS))[::2]) # Too crowded with every bin labelled.
	ax.set_xticklabels(BUCKETS[::2])
	ax.semilogy()

	ax.legend(handles=handles, labels=include_disks)

	plt.show()


	if __name__ == "__main__":
	# Run `zpool iostat -pwvH > output.tsv`.
	with open("output.tsv") as f:
	stats = parse_output(f.read())
	disks = [d for d in stats.keys() if d not in ["tank"]] # Exclude whole-pool stats?
	display_graphs(stats, name="zpool", include_disks=disks)