PrintWindow: screenshot is offset sometimes - winapi

Question

I'm taking screenshots of windows using PrintWindow(). This works finde with the following code.
This works most of the time - I'm calling it very often in a continuos loop - but sometimes it fails. I started saving screenshots of the times it fails. Sometimes they are just black, but sometimes the ClientArea of the captured window is somehow offset and only visible on half the picture. Why is that the case?
EDIT:
Is the PrintWindow function guaranteed to work, even when its called ~100times a second? What makes me wonder is, that after every print window I'm actually verifying two pixels which default colors I stored as a COLORREF at startup. The program says they match almost all the time. Still, once I use GetPixel, the returned value from other pixels is simply black or white from time to time (which should never be the case). Once I save the HDC/HBITMAP as a screenshot (I called GdiFlush before), the screenshot is just plain black (which shouldn't be the case either, as I said, I checked two pixels before...).
Well for now, back to the initial question: Is this definitely my fault, or is it possible that PrintWindow just gives random results from time to time, without returning an error?
Ok, this is the NEW CODE:
/* Try printing the window 10times, in case it fails. */
for (int i = 0; i <= NUM_ITERATIONS_PREP; i++) {
releaseSurface();
prepared_HWND = hwnd;
window_dc = GetWindowDC(hwnd);
surface = CreateCompatibleDC(window_dc);
if (surface) {
RECT r;
GetWindowRect(hwnd, &r);
bmp = CreateCompatibleBitmap(window_dc, r.right - r.left, r.bottom - r.top);
if (!ReleaseDC(hwnd, window_dc)) {
writeLog("ReleaseDC failed.", black, true);
}
if (bmp) {
old_surface_bmp = SelectBitmap(surface, bmp);
//InvalidateRect(hwnd, 0, TRUE);
//UpdateWindow(hwnd);
int result = PrintWindow(hwnd, surface, 0);
if (result != 0) {
if (!verify || verifySurface())
break;
else {
writeLog("Surface could not be verified on iteration " + std::to_string(i) + ". Saving screenshot.", black, true);
saveDebugScreenshot();
QThread::msleep(2);
}
} else {
writeLog("PrintWindow() failed.", black, true);
}
}
else {
writeLog("BMP creation failed.", black, true);
}
}
else {
writeLog("HDC creation failed.", black, true);
}
}
And the new RELEASE CODE:
/* If you call prepareSurface(), call this once the surface isn't need anymore. */
void ClientProfile::releaseSurface() {
SelectObject(surface, old_surface_bmp);
DeleteDC(surface);
DeleteObject(bmp);
prepared_HWND = NULL;
}
releaseSurface() is called either at the beginning of prepareSurface() and additionally every time I'm done processing a table (this means its called once too often for every prepareSurface(). Is that a problem?).
Oh and old_surface_bmp, bmp and surface are declared as private class attributes.