Question

I’m trying to make my data analysis and reports less eye stabbing and more graphical with automatically generated graph-files, and to do this I’ve been playing with matplotlib/pyplot/pylab. Works brilliantly, but when I try to run it on a headless server…

tkinter.TclError: no display name and no $DISPLAY environment variable

For this application I only use PyLab, but after a little google, I added the below to before the pylab import:

import matplotlib
matplotlib.use("Agg")

Which should have changed the backend, but to no effect.

Any ideas what I’m doing wrong?

The remote machine does have X-forwarding capabilities, but since this application shouldn’t NEED to display anything, I believe the usual ssh -X hack is overkill.

Examples:Fiendish Deception

Example working code on same machine

import matplotlib
matplotlib.use("Agg")
import numpy as np
import pylab as pl


xvals=np.arange(100)
yvals=np.cumsum(np.random.random(100))
yvals[-10:]=0
yvals=np.log(yvals)
pl.close()

pl.plot(xvals,yvals)
pl.xlabel("X")
pl.ylabel("Y")
pl.title("Title")

pl.savefig("testgraph.png")

Non-working real code

import matplotlib
matplotlib.use("Agg")
import numpy as np
import pylab as pl
import utility as util
import os

... non graph stuff...
def graph_p(self):
    pl.close()
    channels=range(self.p.shape[0])
    for line in range(self.p.shape[1]):
        yvals=np.ma.masked_invalid(map(util.watts_to_dbmhz,self.p[:,line]))
        pl.plot(channels,yvals) #Error says it occurs here
    pl.xlabel("Subchannel Index")
    pl.ylabel("Power (dbmhz)")
    pl.title("Plot of per-tone power assignments for %d lines"%self.p.shape[1])
    pl.savefig(self.dest+self.scenario+'-power.png')

Nothing else touches pyplot.

Answer 1

Turns out a utility file (not mine!) was pulling in pylab for something else. Shifted the matplotlib backend selection into the initial page.